#!/usr/bin/env python
# -*- coding: utf-8 -*-
# tifffile.py

# Copyright (c) 2008-2017, Christoph Gohlke
# Copyright (c) 2008-2017, The Regents of the University of California
# Produced at the Laboratory for Fluorescence Dynamics
# All rights reserved.
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
# * Redistributions of source code must retain the above copyright
#   notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
#   notice, this list of conditions and the following disclaimer in the
#   documentation and/or other materials provided with the distribution.
# * Neither the name of the copyright holders nor the names of any
#   contributors may be used to endorse or promote products derived
#   from this software without specific prior written permission.

"""Read image and meta data from (bio)TIFF files. Save numpy arrays as TIFF.

Image and metadata can be read from TIFF, BigTIFF, OME-TIFF, STK, LSM, NIH,
SGI, ImageJ, MicroManager, FluoView, SEQ and GEL files.
Only a subset of the TIFF specification is supported, mainly uncompressed
and losslessly compressed 2**(0 to 6) bit integer, 16, 32 and 64-bit float,
grayscale and RGB(A) images, which are commonly used in bio-scientific imaging.
Specifically, reading JPEG and CCITT compressed image data, chroma subsampling,
or EXIF, IPTC, GPS, and XMP metadata is not implemented. Only primary info
records are read for STK, FluoView, MicroManager, and NIH Image formats.

TIFF, the Tagged Image File Format aka Thousands of Incompatible File Formats,
is under the control of Adobe Systems. BigTIFF allows for files greater than
4 GB. STK, LSM, FluoView, SGI, SEQ, GEL, and OME-TIFF, are custom extensions
defined by Molecular Devices (Universal Imaging Corporation), Carl Zeiss
MicroImaging, Olympus, Silicon Graphics International, Media Cybernetics,
Molecular Dynamics, and the Open Microscopy Environment consortium

For command line usage run C{python -m tifffile --help}

  `Christoph Gohlke <http://www.lfd.uci.edu/~gohlke/>`_

  Laboratory for Fluorescence Dynamics, University of California, Irvine

:Version: 2017.06.21.dev

* `CPython 3.6 <http://www.python.org>`_ (64-bit recommended)
* `Numpy 1.11 <http://www.numpy.org>`_
* `Matplotlib 2.0 <http://www.matplotlib.org>`_ (optional for plotting)
* `Tifffile.c 2017.01.10 <http://www.lfd.uci.edu/~gohlke/>`_
  (recommended for faster decoding of PackBits and LZW encoded strings)

2017.06.21.dev (tentative)
    Many backwards incompatible changes improving speed and resource usage:
    Only return main image series for LSM files.
    Add optional movie mode for files with uniform pages.
    Lazy load TIFF pages.
    Add option to not cache TiffPages/Frames loaded from file.
    Use lightweight TiffFrame for IFDs sharing properties with key TiffPage.
    Move module constants to 'CONST' namespace (speed up module import).
    Remove 'pages' parameter from TiffFile.
    Remove TIFFfile alias.
    Deprecate Python 2.
    Remove Record. Return all metadata as dict or numpy.record.
    Add functions to parse STK, MetaSeries, ScanImage, SVS metadata to dicts.
    Use pprint for info strings.
    Fixed reading some UIC tags (bug fix).
    Do not modify input array in imshow  (bug fix).
    Pass 1961 tests.
    Write correct number of sample_format values (bug fix).
    Use Adobe deflate code to write ZIP compressed files.
    Add option to pass tag values as packed binary data for writing.
    Defer tag validation to attribute access.
    Use property instead of lazyattr decorator for simple expressions.
    Write IFDs and tag values on word boundaries.
    Read ScanImage metadata.
    Remove is_rgb and is_indexed attributes from TiffFile.
    Create files used by doctests.
    Read Zeiss SEM metadata.
    Read OME-TIFF with invalid references to external files.
    Rewrite C LZW decoder (5x faster).
    Read corrupted LSM files missing EOI code in LZW stream.
    Add option to append images to existing TIFF files.
    Read files without pages.
    Read S-FEG and Helios NanoLab tags created by FEI software.
    Allow saving Color Filter Array (CFA) images.
    Add info functions returning more information about TiffFile and TiffPage.
    Add option to read specific pages only.
    Remove maxpages argument (backwards incompatible).
    Remove test_tifffile function.
    Pass 1944 tests.
    Improve detection of ImageJ hyperstacks.
    Read TVIPS metadata created by EM-MENU (by Marco Oster).
    Add option to disable using OME-XML metadata.
    Allow non-integer range attributes in modulo tags (by Stuart Berg).
    Do not always memmap contiguous data in page series.
    Add option to specify resolution unit.
    Write grayscale images with extra samples when planarconfig is specified.
    Do not write RGB color images with 2 samples.
    Reorder TiffWriter.save keyword arguments (backwards incompatible).
    Pass 1932 tests.
    TiffWriter, imread, and imsave accept open binary file streams.
    Correctly handle reversed fill order in 2 and 4 bps images (bug fix).
    Implement reverse_bitorder in C.
    Fix saving additional ImageJ metadata.
    Pass 1920 tests.
    Write 8 bytes double tag values using offset if necessary (bug fix).
    Add option to disable writing second image description tag.
    Detect tags with incorrect counts.
    Disable color mapping for LSM.
    Read LSM 6 mosaics.
    Add option to specify directory of memory-mapped files.
    Add command line options to specify vmin and vmax values for colormapping.
    New helper function to apply colormaps.
    Renamed is_palette attributes to is_indexed (backwards incompatible).
    Color-mapped samples are now contiguous (backwards incompatible).
    Do not color-map ImageJ hyperstacks (backwards incompatible).
    Towards supporting Leica SCN.
    Read images with reversed bit order (fill_order is lsb2msb).
    Read RGB OME-TIFF.
    Warn about malformed OME-XML.
    Detect some corrupted ImageJ metadata.
    Better axes labels for 'shaped' files.
    Do not create TiffTag for default values.
    Chroma subsampling is not supported.
    Memory-map data in TiffPageSeries if possible (optional).
    Pass 1906 tests.
    Write ImageJ hyperstacks (optional).
    Read and write LZMA compressed data.
    Specify datetime when saving (optional).
    Save tiled and color-mapped images (optional).
    Ignore void byte_counts and offsets if possible.
    Ignore bogus image_depth tag created by ISS Vista software.
    Decode floating point horizontal differencing (not tiled).
    Save image data contiguously if possible.
    Only read first IFD from ImageJ files if possible.
    Read ImageJ 'raw' format (files larger than 4 GB).
    TiffPageSeries class for pages with compatible shape and data type.
    Try to read incomplete tiles.
    Open file dialog if no filename is passed on command line.
    Ignore errors when decoding OME-XML.
    Rename decoder functions (backwards incompatible)
    TiffWriter class for incremental writing images.
    Simplify examples.
    Add memmap function to FileHandle.
    Add function to determine if image data in TiffPage is memory-mappable.
    Do not close files if multifile_close parameter is False.
    Pass 1730 tests.
    Return all extrasamples by default (backwards incompatible).
    Read data from series of pages into memory-mapped array (optional).
    Squeeze OME dimensions (backwards incompatible).
    Workaround missing EOI code in strips.
    Support image and tile depth tags (SGI extension).
    Better handling of STK/UIC tags (backwards incompatible).
    Disable color mapping for STK.
    Julian to datetime converter.
    TIFF ASCII type may be NULL separated.
    Unwrap strip offsets for LSM files greater than 4 GB.
    Correct strip byte counts in compressed LSM files.
    Skip missing files in OME series.
    Read embedded TIFF files.
    Save rational numbers as type 5 (bug fix).
    Keep other files in OME multi-file series closed.
    FileHandle class to abstract binary file handle.
    Disable color mapping for bad OME-TIFF produced by bio-formats.
    Read bad OME-XML produced by ImageJ when cropping.
    Allow zlib compress data in imsave function (optional).
    Memory-map contiguous image data (optional).
    Read MicroManager metadata and little endian ImageJ tag.
    Save extra tags in imsave function.
    Save tags in ascending order by code (bug fix).
    Accept file like objects (read from OIB files).
    Rename TIFFfile to TiffFile and TIFFpage to TiffPage.
    TiffSequence class for reading sequence of TIFF files.
    Read UltraQuant tags.
    Allow float numbers as resolution in imsave function.
    Read MD GEL tags and NIH Image header.
    Read ImageJ tags.

The API is not stable yet and might change between revisions.

Tested on little-endian platforms only.

TIFF trees defined via sub_ifds tags are not supported.

Other Python packages and modules for reading bio-scientific TIFF files:

*  `python-bioformats <https://github.com/CellProfiler/python-bioformats>`_
*  `Imread <https://github.com/luispedro/imread>`_
*  `PyLibTiff <https://github.com/pearu/pylibtiff>`_
*  `SimpleITK <http://www.simpleitk.org>`_
*  `PyLSM <https://launchpad.net/pylsm>`_
*  `PyMca.TiffIO.py <https://github.com/vasole/pymca>`_ (same as fabio.TiffIO)
*  `BioImageXD.Readers <http://www.bioimagexd.net/>`_
*  `Cellcognition.io <http://cellcognition.org/>`_
*  `pymimage <https://github.com/ardoi/pymimage>`_

*   Egor Zindy, University of Manchester, for cz_lsm_scan_info specifics.
*   Wim Lewis for a bug fix and some read_cz_lsm functions.
*   Hadrien Mary for help on reading MicroManager files.
*   Christian Kliche for help writing tiled and color-mapped files.

(1) TIFF 6.0 Specification and Supplements. Adobe Systems Incorporated.
(2) TIFF File Format FAQ. http://www.awaresystems.be/imaging/tiff/faq.html
(3) MetaMorph Stack (STK) Image File Format.
(4) Image File Format Description LSM 5/7 Release 6.0 (ZEN 2010).
    Carl Zeiss MicroImaging GmbH. BioSciences. May 10, 2011
(5) The OME-TIFF format.
(6) UltraQuant(r) Version 6.0 for Windows Start-Up Guide.
(7) Micro-Manager File Formats.
(8) Tags for TIFF and Related Specifications. Digital Preservation.
(9) ScanImage BigTiff Specification - ScanImage 2016.

>>> data = numpy.random.rand(5, 301, 219)
>>> imsave('temp.tif', data)

>>> image = imread('temp.tif')
>>> numpy.testing.assert_array_equal(image, data)

>>> with TiffFile('temp.tif') as tif:
...     images = tif.asarray()
...     for page in tif.pages:
...         for tag in page.tags.values():
...             _ = tag.name, tag.value
...         image = page.asarray()


from __future__ import division, print_function

import sys
import os
import re
import glob
import math
import zlib
import time
import json
import struct
import warnings
import tempfile
import datetime
import collections
# from fractions import Fraction  # delay import
# from xml.etree import cElementTree as etree  # delay import

import numpy

    import lzma
except ImportError:
        import backports.lzma as lzma
    except ImportError:
        lzma = None

__version__ = '2017.06.21.dev'
__docformat__ = 'restructuredtext en'
__all__ = (
    'imsave', 'imread', 'imshow', 'TiffFile', 'TiffWriter', 'TiffSequence',
    # utility functions used in oiffile and czifile
    'FileHandle', 'lazyattr', 'natural_sorted', 'decode_lzw', 'stripnull')

def imsave(file, data, **kwargs):
    """Write image data to TIFF file.

    Refer to the TiffWriter class and member functions for documentation.

    file : str or binary stream
        File name or writable binary stream, such as a open file or BytesIO.
    data : array_like
        Input image. The last dimensions are assumed to be image depth,
        height, width, and samples.
    kwargs : dict
        Parameters 'append', 'byteorder', 'bigtiff', 'software', and 'imagej',
        are passed to the TiffWriter class.
        Parameters 'photometric', 'planarconfig', 'resolution', 'compress',
        'colormap', 'tile', 'description', 'datetime', 'metadata', 'contiguous'
        and 'extratags' are passed to the TiffWriter.save function.

    >>> data = numpy.random.rand(2, 5, 3, 301, 219)
    >>> imsave('temp.tif', data, compress=6, metadata={'axes': 'TZCYX'})

    tifargs = parse_kwargs(kwargs, 'append', 'bigtiff', 'byteorder',
                           'software', 'imagej')

    if 'bigtiff' not in tifargs and 'imagej' not in tifargs and (
            data.size*data.dtype.itemsize > 2000*2**20):
        tifargs['bigtiff'] = True

    with TiffWriter(file, **tifargs) as tif:
        tif.save(data, **kwargs)

def imread(files, **kwargs):
    """Return image data from TIFF file(s) as numpy array.

    Refer to the TiffFile class and member functions for documentation.

    files : str, binary stream, or sequence
        File name, seekable binary stream, glob pattern, or sequence of
        file names.
    kwargs : dict
        Parameters 'multifile', 'multifile_close', 'fastij', and 'is_ome'
        are passed to the TiffFile class.
        The 'pattern' parameter is passed to the TiffSequence class.
        Other parameters are passed to the asarray functions.
        The first image series is returned if no arguments are provided.

    >>> imsave('temp.tif', numpy.random.rand(3, 4, 301, 219))
    >>> im = imread('temp.tif', key=0)
    >>> im.shape
    (4, 301, 219)
    >>> ims = imread(['temp.tif', 'temp.tif'])
    >>> ims.shape
    (2, 3, 4, 301, 219)

    kwargs_file = parse_kwargs(kwargs, 'multifile', 'multifile_close',
                               'fastij', 'is_ome')
    kwargs_seq = parse_kwargs(kwargs, 'pattern')

    if isinstance(files, basestring) and any(i in files for i in '?*'):
        files = glob.glob(files)
    if not files:
        raise ValueError('no files found')
    if not hasattr(files, 'seek') and len(files) == 1:
        files = files[0]

    if isinstance(files, basestring) or hasattr(files, 'seek'):
        with TiffFile(files, **kwargs_file) as tif:
            return tif.asarray(**kwargs)
        with TiffSequence(files, **kwargs_seq) as imseq:
            return imseq.asarray(**kwargs)

class lazyattr(object):
    """Attribute whose value is computed on first access."""
    __slots__ = ('func',)

    def __init__(self, func):
        self.func = func
        # self.lock = threading.RLock()

    def __get__(self, instance, owner):
        # with self.lock:
        if instance is None:
            return self
        value = self.func(instance)
        if value is NotImplemented:
            return getattr(super(owner, instance), self.func.__name__)
        setattr(instance, self.func.__name__, value)
        return value

class TiffWriter(object):
    """Write image data to TIFF file.

    TiffWriter instances must be closed using the 'close' method, which is
    automatically called when using the 'with' context manager.

    >>> data = numpy.random.rand(2, 5, 3, 301, 219)
    >>> with TiffWriter('temp.tif', bigtiff=True) as tif:
    ...     for i in range(data.shape[0]):
    ...         tif.save(data[i], compress=6)

    def __init__(self, file, append=False, bigtiff=False, byteorder=None,
                 software='tifffile.py', imagej=False):
        """Open a TIFF file for writing.

        Existing files are overwritten by default.
        Use bigtiff=True when creating files larger than 2 GB.

        file : str, binary stream, or FileHandle
            File name or writable binary stream, such as a open file
            or BytesIO.
            The file is created if it does not exist.
        append : bool
            If True and 'file' is an existing standard TIFF file, image data
            and tags are appended to the file.
            Appending data may corrupt specifically formatted TIFF files
            such as LSM, STK, ImageJ, NIH, or FluoView.
        bigtiff : bool
            If True, the BigTIFF format is used.
        byteorder : {'<', '>'}
            The endianness of the data in the file.
            By default this is the system's native byte order.
        software : str
            Name of the software used to create the file.
            Saved with the first page in the file only.
        imagej : bool
            If True, write an ImageJ hyperstack compatible file.
            This format can handle data types uint8, uint16, or float32 and
            data shapes up to 6 dimensions in TZCYXS order.
            RGB images (S=3 or S=4) must be uint8.
            ImageJ's default byte order is big endian but this implementation
            uses the system's native byte order by default.
            ImageJ does not support BigTIFF format or LZMA compression.
            The ImageJ file format is undocumented.

        if append:
            # determine if file is an existing TIFF file that can be extended
                with FileHandle(file, mode='rb', size=0) as fh:
                    pos = fh.tell()
                        with TiffFile(fh) as tif:
                            if (append != 'force' and
                                any(getattr(tif, 'is_'+a) for a in
                                    ('lsm', 'stk', 'imagej', 'nih', 'fluoview',
                                raise ValueError("contains metadata")
                            byteorder = tif.byteorder
                            bigtiff = tif.is_bigtiff
                            imagej = tif.is_imagej
                            self._ifd_offset = tif.pages.next_page_offset
                            if tif.pages:
                                software = None
                    except Exception as e:
                        raise ValueError("can not append to file: %s" % str(e))
            except (IOError, FileNotFoundError):
                append = False

        if byteorder not in (None, '<', '>'):
            raise ValueError("invalid byteorder %s" % byteorder)
        if byteorder is None:
            byteorder = '<' if sys.byteorder == 'little' else '>'
        if imagej and bigtiff:
            warnings.warn("writing incompatible bigtiff ImageJ")

        self._byteorder = byteorder
        self._software = software
        self._imagej = bool(imagej)
        self._metadata = None
        self._colormap = None

        self._description_offset = 0
        self._description_len_offset = 0
        self._description_len = 0

        self._tags = None
        self._shape = None  # normalized shape of data in consecutive pages
        self._data_shape = None  # shape of data in consecutive pages
        self._data_dtype = None  # data type
        self._data_offset = None  # offset to data
        self._data_byte_counts = None  # byte counts per plane
        self._tag_offsets = None  # strip or tile offset tag code

        if bigtiff:
            self._bigtiff = True
            self._offset_size = 8
            self._tag_size = 20
            self._tagno_format = 'Q'
            self._offset_format = 'Q'
            self._value_format = '8s'
            self._bigtiff = False
            self._offset_size = 4
            self._tag_size = 12
            self._tagno_format = 'H'
            self._offset_format = 'I'
            self._value_format = '4s'

        if append:
            self._fh = FileHandle(file, mode='r+b', size=0)
            self._fh.seek(0, 2)
            self._fh = FileHandle(file, mode='wb', size=0)
            self._fh.write({'<': b'II', '>': b'MM'}[byteorder])
            if bigtiff:
                self._fh.write(struct.pack(byteorder+'HHH', 43, 8, 0))
                self._fh.write(struct.pack(byteorder+'H', 42))
            # first IFD
            self._ifd_offset = self._fh.tell()
            self._fh.write(struct.pack(byteorder+self._offset_format, 0))

    def save(self, data, photometric=None, planarconfig=None, tile=None,
             contiguous=True, compress=0, colormap=None,
             description=None, datetime=None, resolution=None,
             metadata={}, extratags=()):
        """Write image data and tags to TIFF file.

        Image data are written in one stripe per plane by default.
        Dimensions larger than 2 to 4 (depending on photometric mode, planar
        configuration, and SGI mode) are flattened and saved as separate pages.
        The 'sample_format' and 'bits_per_sample' tags are derived from
        the data type.

        data : numpy.ndarray
            Input image. The last dimensions are assumed to be image depth,
            height (length), width, and samples.
            If a colormap is provided, the dtype must be uint8 or uint16 and
            the data values are indices into the last dimension of the
        photometric : {'minisblack', 'miniswhite', 'rgb', 'palette', 'cfa'}
            The color space of the image data.
            By default this setting is inferred from the data shape and the
            value of colormap.
            For CFA images, DNG tags must be specified in extratags.
        planarconfig : {'contig', 'planar'}
            Specifies if samples are stored contiguous or in separate planes.
            By default this setting is inferred from the data shape.
            If this parameter is set, extra samples are used to store grayscale
            'contig': last dimension contains samples.
            'planar': third last dimension contains samples.
        tile : tuple of int
            The shape (depth, length, width) of image tiles to write.
            If None (default), image data are written in one stripe per plane.
            The tile length and width must be a multiple of 16.
            If the tile depth is provided, the SGI image_depth and tile_depth
            tags are used to save volume data. Few software can read the
            SGI format, e.g. MeVisLab.
        contiguous : bool
            If True (default) and the data and parameters are compatible with
            previous ones, if any, the data are stored contiguously after
            the previous one. Parameters 'photometric' and 'planarconfig' are
        compress : int or 'lzma'
            Values from 0 to 9 controlling the level of zlib compression.
            If 0, data are written uncompressed (default).
            Compression cannot be used to write contiguous files.
            If 'lzma', LZMA compression is used, which is not available on
            all platforms.
        colormap : numpy.ndarray
            RGB color values for the corresponding data value.
            Must be of shape (3, 2**(data.itemsize*8)) and dtype uint16.
        description : str
            The subject of the image. Saved with the first page only.
            Cannot be used with the ImageJ format.
        datetime : datetime
            Date and time of image creation. Saved with the first page only.
            If None (default), the current date and time is used.
        resolution : (float, float[, str]) or ((int, int), (int, int)[, str])
            X and Y resolutions in pixels per resolution unit as float or
            rational numbers.
            A third, optional parameter specifies the resolution unit,
            which must be None (default for ImageJ), 'inch' (default), or 'cm'.
        metadata : dict
            Additional meta data to be saved along with shape information
            in JSON or ImageJ formats in an image_description tag.
            If None, do not write a second image_description tag.
        extratags : sequence of tuples
            Additional tags as [(code, dtype, count, value, writeonce)].

            code : int
                The TIFF tag Id.
            dtype : str
                Data type of items in 'value' in Python struct format.
                One of B, s, H, I, 2I, b, h, i, 2i, f, d, Q, or q.
            count : int
                Number of data values. Not used for string or byte string
            value : sequence
                'Count' values compatible with 'dtype'.
                Byte strings must contain count values of dtype packed as
                binary data.
            writeonce : bool
                If True, the tag is written to the first page only.

        # TODO: refactor this function
        fh = self._fh
        byteorder = self._byteorder
        tagno_format = self._tagno_format
        value_format = self._value_format
        offset_format = self._offset_format
        offset_size = self._offset_size
        tag_size = self._tag_size

        data = numpy.asarray(data, dtype=byteorder+data.dtype.char, order='C')
        if data.size == 0:
            raise ValueError("can not save empty array")

        # just append contiguous data if possible
        if self._data_shape:
            if (not contiguous or
                    self._data_shape[1:] != data.shape or
                    self._data_dtype != data.dtype or
                    (compress and self._tags) or
                    tile or
                    not numpy.array_equal(colormap, self._colormap)):
                # incompatible shape, dtype, compression mode, or colormap
                self._description_offset = 0
                self._description_len_offset = 0
                self._data_shape = None
                self._colormap = None
                if self._imagej:
                    raise ValueError(
                        "ImageJ does not support non-contiguous data")
                # consecutive mode
                self._data_shape = (self._data_shape[0] + 1,) + data.shape
                if not compress:
                    # write contiguous data, write ifds/tags later

        if photometric not in (None, 'minisblack', 'miniswhite',
                               'rgb', 'palette', 'cfa'):
            raise ValueError("invalid photometric %s" % photometric)
        if planarconfig not in (None, 'contig', 'planar'):
            raise ValueError("invalid planarconfig %s" % planarconfig)

        # prepare compression
        if not compress:
            compress = False
            compress_tag = 1
        elif compress == 'lzma':
            compress = lzma.compress
            compress_tag = 34925
            if self._imagej:
                raise ValueError("ImageJ can not handle LZMA compression")
        elif not 0 <= compress <= 9:
            raise ValueError("invalid compression level %s" % compress)
        elif compress:
            def compress(data, level=compress):
                return zlib.compress(data, level)
            compress_tag = 8

        # prepare ImageJ format
        if self._imagej:
            if description:
                warnings.warn("not writing description to ImageJ file")
                description = None
            volume = False
            if data.dtype.char not in 'BHhf':
                raise ValueError("ImageJ does not support data type '%s'"
                                 % data.dtype.char)
            ijrgb = photometric == 'rgb' if photometric else None
            if data.dtype.char not in 'B':
                ijrgb = False
            ijshape = imagej_shape(data.shape, ijrgb)
            if ijshape[-1] in (3, 4):
                photometric = 'rgb'
                if data.dtype.char not in 'B':
                    raise ValueError("ImageJ does not support data type '%s' "
                                     "for RGB" % data.dtype.char)
            elif photometric is None:
                photometric = 'minisblack'
                planarconfig = None
            if planarconfig == 'planar':
                raise ValueError("ImageJ does not support planar images")
                planarconfig = 'contig' if ijrgb else None

        # verify colormap and indices
        if colormap is not None:
            if data.dtype.char not in 'BH':
                raise ValueError("invalid data dtype for palette mode")
            colormap = numpy.asarray(colormap, dtype=byteorder+'H')
            if colormap.shape != (3, 2**(data.itemsize * 8)):
                raise ValueError("invalid color map shape")
            self._colormap = colormap

        # verify tile shape
        if tile:
            tile = tuple(int(i) for i in tile[:3])
            volume = len(tile) == 3
            if (len(tile) < 2 or tile[-1] % 16 or tile[-2] % 16 or
                    any(i < 1 for i in tile)):
                raise ValueError("invalid tile shape")
            tile = ()
            volume = False

        # normalize data shape to 5D or 6D, depending on volume:
        #   (pages, planar_samples, [depth,] height, width, contig_samples)
        data_shape = data.shape

        if photometric == 'rgb':
            data = reshape_nd(data, 3)
            data = reshape_nd(data, 2)

        shape = data.shape

        samplesperpixel = 1
        extrasamples = 0
        if volume and data.ndim < 3:
            volume = False
        if colormap is not None:
            photometric = 'palette'
            planarconfig = None
        if photometric is None:
            photometric = 'minisblack'
            if planarconfig == 'contig':
                if data.ndim > 2 and shape[-1] in (3, 4):
                    photometric = 'rgb'
            elif planarconfig == 'planar':
                if volume and data.ndim > 3 and shape[-4] in (3, 4):
                    photometric = 'rgb'
                elif data.ndim > 2 and shape[-3] in (3, 4):
                    photometric = 'rgb'
            elif data.ndim > 2 and shape[-1] in (3, 4):
                photometric = 'rgb'
            elif self._imagej:
                photometric = 'minisblack'
            elif volume and data.ndim > 3 and shape[-4] in (3, 4):
                photometric = 'rgb'
            elif data.ndim > 2 and shape[-3] in (3, 4):
                photometric = 'rgb'
        if planarconfig and len(shape) <= (3 if volume else 2):
            planarconfig = None
            photometric = 'minisblack'
        if photometric == 'rgb':
            if len(shape) < 3:
                raise ValueError("not a RGB(A) image")
            if len(shape) < 4:
                volume = False
            if planarconfig is None:
                if shape[-1] in (3, 4):
                    planarconfig = 'contig'
                elif shape[-4 if volume else -3] in (3, 4):
                    planarconfig = 'planar'
                elif shape[-1] > shape[-4 if volume else -3]:
                    planarconfig = 'planar'
                    planarconfig = 'contig'
            if planarconfig == 'contig':
                data = data.reshape((-1, 1) + shape[(-4 if volume else -3):])
                samplesperpixel = data.shape[-1]
                data = data.reshape(
                    (-1,) + shape[(-4 if volume else -3):] + (1,))
                samplesperpixel = data.shape[1]
            if samplesperpixel > 3:
                extrasamples = samplesperpixel - 3
        elif photometric == 'cfa':
            if len(shape) != 2:
                raise ValueError("invalid CFA image")
            volume = False
            planarconfig = None
            data = data.reshape((-1, 1) + shape[-2:] + (1,))
            if 50706 not in (et[0] for et in extratags):
                raise ValueError("must specify DNG tags for CFA image")
        elif planarconfig and len(shape) > (3 if volume else 2):
            if planarconfig == 'contig':
                data = data.reshape((-1, 1) + shape[(-4 if volume else -3):])
                samplesperpixel = data.shape[-1]
                data = data.reshape(
                    (-1,) + shape[(-4 if volume else -3):] + (1,))
                samplesperpixel = data.shape[1]
            extrasamples = samplesperpixel - 1
            planarconfig = None
            # remove trailing 1s
            while len(shape) > 2 and shape[-1] == 1:
                shape = shape[:-1]
            if len(shape) < 3:
                volume = False
            data = data.reshape(
                (-1, 1) + shape[(-3 if volume else -2):] + (1,))

        # normalize shape to 6D
        assert len(data.shape) in (5, 6)
        if len(data.shape) == 5:
            data = data.reshape(data.shape[:2] + (1,) + data.shape[2:])
        shape = data.shape

        if tile and not volume:
            tile = (1, tile[-2], tile[-1])

        if photometric == 'palette':
            if (samplesperpixel != 1 or extrasamples or
                    shape[1] != 1 or shape[-1] != 1):
                raise ValueError("invalid data shape for palette mode")

        if photometric == 'rgb' and samplesperpixel == 2:
            raise ValueError("not a RGB image (samplesperpixel=2)")

        bytestr = bytes if sys.version[0] == '2' else (
            lambda x: bytes(x, 'utf-8') if isinstance(x, str) else x)
        tags = []  # list of (code, ifdentry, ifdvalue, writeonce)

        strip_or_tile = 'tile' if tile else 'strip'
        tag_byte_counts = CONST.TIFF_TAGS_NAMES[strip_or_tile + '_byte_counts']
        tag_offsets = CONST.TIFF_TAGS_NAMES[strip_or_tile + '_offsets']
        self._tag_offsets = tag_offsets

        def pack(fmt, *val):
            return struct.pack(byteorder+fmt, *val)

        def addtag(code, dtype, count, value, writeonce=False):
            # Compute ifdentry & ifdvalue bytes from code, dtype, count, value
            # Append (code, ifdentry, ifdvalue, writeonce) to tags list
            code = int(CONST.TIFF_TAGS_NAMES.get(code, code))
                tifftype = CONST.TIFF_DATA_DTYPES[dtype]
            except KeyError:
                raise ValueError("unknown dtype %s" % dtype)
            rawcount = count

            if dtype == 's':
                # strings
                value = bytestr(value) + b'\0'
                count = rawcount = len(value)
                rawcount = value.find(b'\0\0')
                if rawcount < 0:
                    rawcount = count
                    rawcount += 1  # length of string without buffer
                value = (value,)
            elif isinstance(value, bytes):
                # packed binary data
                dtsize = struct.calcsize(dtype)
                if len(value) % dtsize:
                    raise ValueError('invalid packed binary data')
                count = len(value) // dtsize
            if len(dtype) > 1:
                count *= int(dtype[:-1])
                dtype = dtype[-1]
            ifdentry = [pack('HH', code, tifftype),
                        pack(offset_format, rawcount)]
            ifdvalue = None
            if struct.calcsize(dtype) * count <= offset_size:
                # value(s) can be written directly
                if isinstance(value, bytes):
                    ifdentry.append(pack(value_format, value))
                elif count == 1:
                    if isinstance(value, (tuple, list, numpy.ndarray)):
                        value = value[0]
                    ifdentry.append(pack(value_format, pack(dtype, value)))
                                         pack(str(count)+dtype, *value)))
                # use offset to value(s)
                ifdentry.append(pack(offset_format, 0))
                if isinstance(value, bytes):
                    ifdvalue = value
                elif isinstance(value, numpy.ndarray):
                    assert value.size == count
                    assert value.dtype.char == dtype
                    ifdvalue = value.tostring()
                elif isinstance(value, (tuple, list)):
                    ifdvalue = pack(str(count)+dtype, *value)
                    ifdvalue = pack(dtype, value)
            tags.append((code, b''.join(ifdentry), ifdvalue, writeonce))

        def rational(arg, max_denominator=1000000):
            # return nominator and denominator from float or two integers
            from fractions import Fraction  # delayed import
                f = Fraction.from_float(arg)
            except TypeError:
                f = Fraction(arg[0], arg[1])
            f = f.limit_denominator(max_denominator)
            return f.numerator, f.denominator

        if description:
            # user provided description
            addtag('image_description', 's', 0, description, writeonce=True)

        # write shape and metadata to image_description
        self._metadata = {} if not metadata else metadata
        if self._imagej:
            description = imagej_description(
                data_shape, shape[-1] in (3, 4), self._colormap is not None,
        elif metadata or metadata == {}:
            description = json_description(
                data_shape, self._colormap is not None, **self._metadata)
            description = None
        if description:
            # add 32 bytes buffer
            # the image description might be updated later with the final shape
            description += b'\0'*32
            self._description_len = len(description)
            addtag('image_description', 's', 0, description, writeonce=True)

        if self._software:
            addtag('software', 's', 0, self._software, writeonce=True)
            self._software = None  # only save to first page in file
        if datetime is None:
            datetime = self._now()
        addtag('datetime', 's', 0, datetime.strftime("%Y:%m:%d %H:%M:%S"),
        addtag('compression', 'H', 1, compress_tag)
        addtag('image_width', 'I', 1, shape[-2])
        addtag('image_length', 'I', 1, shape[-3])
        if tile:
            addtag('tile_width', 'I', 1, tile[-1])
            addtag('tile_length', 'I', 1, tile[-2])
            if tile[0] > 1:
                addtag('image_depth', 'I', 1, shape[-4])
                addtag('tile_depth', 'I', 1, tile[0])
        addtag('new_subfile_type', 'I', 1, 0)
        sampleformat = {'u': 1, 'i': 2, 'f': 3, 'c': 6}[data.dtype.kind]
        addtag('sample_format', 'H', samplesperpixel,
               (sampleformat,) * samplesperpixel)
        addtag('photometric', 'H', 1, {'miniswhite': 0, 'minisblack': 1,
                                       'rgb': 2, 'palette': 3,
                                       'cfa': 32803}[photometric])
        if colormap is not None:
            addtag('color_map', 'H', colormap.size, colormap)
        addtag('samples_per_pixel', 'H', 1, samplesperpixel)
        if planarconfig and samplesperpixel > 1:
            addtag('planar_configuration', 'H', 1, 1
                   if planarconfig == 'contig' else 2)
            addtag('bits_per_sample', 'H', samplesperpixel,
                   (data.dtype.itemsize * 8,) * samplesperpixel)
            addtag('bits_per_sample', 'H', 1, data.dtype.itemsize * 8)
        if extrasamples:
            if photometric == 'rgb' and extrasamples == 1:
                addtag('extra_samples', 'H', 1, 1)  # associated alpha channel
                addtag('extra_samples', 'H', extrasamples, (0,) * extrasamples)
        if resolution:
            addtag('x_resolution', '2I', 1, rational(resolution[0]))
            addtag('y_resolution', '2I', 1, rational(resolution[1]))
            if len(resolution) > 2:
                resolution_unit = {None: 1, 'inch': 2, 'cm': 3}[resolution[2]]
            elif self._imagej:
                resolution_unit = 1
                resolution_unit = 2
            addtag('resolution_unit', 'H', 1, resolution_unit)
        if not tile:
            addtag('rows_per_strip', 'I', 1, shape[-3])  # * shape[-4]

        if tile:
            # use one chunk per tile per plane
            tiles = ((shape[2] + tile[0] - 1) // tile[0],
                     (shape[3] + tile[1] - 1) // tile[1],
                     (shape[4] + tile[2] - 1) // tile[2])
            numtiles = product(tiles) * shape[1]
            strip_byte_counts = [
                product(tile) * shape[-1] * data.dtype.itemsize] * numtiles
            addtag(tag_byte_counts, offset_format, numtiles, strip_byte_counts)
            addtag(tag_offsets, offset_format, numtiles, [0] * numtiles)
            # allocate tile buffer
            chunk = numpy.empty(tile + (shape[-1],), dtype=data.dtype)
            # use one strip per plane
            strip_byte_counts = [
                data[0, 0].size * data.dtype.itemsize] * shape[1]
            addtag(tag_byte_counts, offset_format, shape[1], strip_byte_counts)
            addtag(tag_offsets, offset_format, shape[1], [0] * shape[1])

        # add extra tags from user
        for t in extratags:

        # TODO: check TIFFReadDirectoryCheckOrder warning in files containing
        #   multiple tags of same code
        # the entries in an IFD must be sorted in ascending order by tag code
        tags = sorted(tags, key=lambda x: x[0])

        if not (self._bigtiff or self._imagej) and (
                fh.tell() + data.size*data.dtype.itemsize > 2**31-1):
            raise ValueError("data too large for standard TIFF file")

        # if not compressed or tiled, write the first ifd and then all data
        # contiguously; else, write all ifds and data interleaved
        for pageindex in range(shape[0] if (compress or tile) else 1):
            # update pointer at ifd_offset
            pos = fh.tell()
            if pos % 2:
                # location of IFD must begin on a word boundary
                pos += 1
            fh.write(pack(offset_format, pos))

            # write ifdentries
            fh.write(pack(tagno_format, len(tags)))
            tag_offset = fh.tell()
            fh.write(b''.join(t[1] for t in tags))
            self._ifd_offset = fh.tell()
            fh.write(pack(offset_format, 0))  # offset to next IFD

            # write tag values and patch offsets in ifdentries, if necessary
            for tagindex, tag in enumerate(tags):
                if tag[2]:
                    pos = fh.tell()
                    if pos % 2:
                        # tag value is expected to begin on a word boundary
                        pos += 1
                    fh.seek(tag_offset + tagindex*tag_size + offset_size + 4)
                    fh.write(pack(offset_format, pos))
                    if tag[0] == tag_offsets:
                        strip_offsets_offset = pos
                    elif tag[0] == tag_byte_counts:
                        strip_byte_counts_offset = pos
                    elif tag[0] == 270 and tag[2].endswith(b'\0\0\0\0'):
                        # image description buffer
                        self._description_offset = pos
                        self._description_len_offset = (
                            tag_offset + tagindex * tag_size + 4)

            # write image data
            data_offset = fh.tell()
            if compress:
                strip_byte_counts = []
            if tile:
                for plane in data[pageindex]:
                    for tz in range(tiles[0]):
                        for ty in range(tiles[1]):
                            for tx in range(tiles[2]):
                                c0 = min(tile[0], shape[2] - tz*tile[0])
                                c1 = min(tile[1], shape[3] - ty*tile[1])
                                c2 = min(tile[2], shape[4] - tx*tile[2])
                                chunk[c0:, c1:, c2:] = 0
                                chunk[:c0, :c1, :c2] = plane[
                                if compress:
                                    t = compress(chunk)
            elif compress:
                for plane in data[pageindex]:
                    plane = compress(plane)

            # update strip/tile offsets and byte_counts if necessary
            pos = fh.tell()
            for tagindex, tag in enumerate(tags):
                if tag[0] == tag_offsets:  # strip/tile offsets
                    if tag[2]:
                        strip_offset = data_offset
                        for size in strip_byte_counts:
                            fh.write(pack(offset_format, strip_offset))
                            strip_offset += size
                        fh.seek(tag_offset + tagindex*tag_size +
                                offset_size + 4)
                        fh.write(pack(offset_format, data_offset))
                elif tag[0] == tag_byte_counts:  # strip/tile byte_counts
                    if compress:
                        if tag[2]:
                            for size in strip_byte_counts:
                                fh.write(pack(offset_format, size))
                            fh.seek(tag_offset + tagindex*tag_size +
                                    offset_size + 4)
                            fh.write(pack(offset_format, strip_byte_counts[0]))

            # remove tags that should be written only once
            if pageindex == 0:
                tags = [tag for tag in tags if not tag[-1]]

        # if uncompressed, write remaining ifds/tags later
        if not (compress or tile):
            self._tags = tags

        self._shape = shape
        self._data_shape = (1,) + data_shape
        self._data_dtype = data.dtype
        self._data_offset = data_offset
        self._data_byte_counts = strip_byte_counts

    def _write_remaining_pages(self):
        """Write outstanding IFDs and tags to file."""
        if not self._tags:

        fh = self._fh
        byteorder = self._byteorder
        tagno_format = self._tagno_format
        offset_format = self._offset_format
        offset_size = self._offset_size
        tag_size = self._tag_size
        data_offset = self._data_offset
        page_data_size = sum(self._data_byte_counts)
        tag_bytes = b''.join(t[1] for t in self._tags)
        pageno = self._shape[0] * self._data_shape[0] - 1

        pos = fh.tell()
        if not self._bigtiff and pos + len(tag_bytes) * pageno > 2**32 - 256:
            if self._imagej:
                warnings.warn("truncating ImageJ file")
            raise ValueError("data too large for non-bigtiff file")

        def pack(fmt, *val):
            return struct.pack(byteorder+fmt, *val)

        for _ in range(pageno):
            # update pointer at ifd_offset
            pos = fh.tell()
            if pos % 2:
                # location of IFD must begin on a word boundary
                pos += 1
            fh.write(pack(offset_format, pos))

            # write ifd entries
            fh.write(pack(tagno_format, len(self._tags)))
            tag_offset = fh.tell()
            self._ifd_offset = fh.tell()
            fh.write(pack(offset_format, 0))  # offset to next IFD

            # offset to image data
            data_offset += page_data_size

            # write tag values and patch offsets in ifdentries, if necessary
            for tagindex, tag in enumerate(self._tags):
                if tag[2]:
                    pos = fh.tell()
                    if pos % 2:
                        # tag value is expected to begin on a word boundary
                        pos += 1
                    fh.seek(tag_offset + tagindex*tag_size + offset_size + 4)
                    fh.write(pack(offset_format, pos))
                    if tag[0] == self._tag_offsets:
                        strip_offsets_offset = pos

            # update strip/tile offsets if necessary
            pos = fh.tell()
            for tagindex, tag in enumerate(self._tags):
                if tag[0] == self._tag_offsets:  # strip/tile offsets
                    if tag[2]:
                        strip_offset = data_offset
                        for size in self._data_byte_counts:
                            fh.write(pack(offset_format, strip_offset))
                            strip_offset += size
                        fh.seek(tag_offset + tagindex*tag_size +
                                offset_size + 4)
                        fh.write(pack(offset_format, data_offset))

        self._tags = None
        self._data_dtype = None
        self._data_offset = None
        self._data_byte_counts = None
        # do not reset _shape or _data_shape

    def _write_image_description(self):
        """Write meta data to image_description tag."""
        if (not self._data_shape or self._data_shape[0] == 1 or
                self._description_offset <= 0):

        colormapped = self._colormap is not None
        if self._imagej:
            isrgb = self._shape[-1] in (3, 4)
            description = imagej_description(
                self._data_shape, isrgb, colormapped, **self._metadata)
            description = json_description(
                self._data_shape, colormapped, **self._metadata)

        # rewrite description and its length to file
        description = description[:self._description_len-1]
        pos = self._fh.tell()

        self._description_offset = 0
        self._description_len_offset = 0
        self._description_len = 0

    def _now(self):
        """Return current date and time."""
        return datetime.datetime.now()

    def close(self, truncate=False):
        """Write remaining pages (if not truncate) and close file handle."""
        if not truncate:

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_value, traceback):

class TiffFile(object):
    """Read image and metadata from TIFF, STK, LSM, and FluoView files.

    TiffFile instances must be closed using the 'close' method, which is
    automatically called when using the 'with' context manager.

    pages : TiffPages
        Sequence of TIFF pages read from file.
    series : list of TiffPageSeries
        TIFF pages with compatible shapes and types.
    byteorder : '>', '<'
        The endianness of data in the file.
        '>': big-endian (Motorola). '>': little-endian (Intel).

    All attributes are read-only.

    >>> imsave('temp.tif', numpy.random.rand(5, 301, 219))
    >>> with TiffFile('temp.tif') as tif:
    ...     data = tif.asarray()
    ...     data.shape
    (5, 301, 219)

    def __init__(self, arg, name=None, offset=None, size=None,
                 multifile=True, multifile_close=True,
                 fastij=True, movie=None, **kwargs):
        """Initialize instance from file.

        arg : str or open file
            Name of file or open file object.
            The file objects are closed in TiffFile.close().
        name : str
            Optional name of file in case 'arg' is a file handle.
        offset : int
            Optional start position of embedded file. By default this is
            the current file position.
        size : int
            Optional size of embedded file. By default this is the number
            of bytes from the 'offset' to the end of the file.
        multifile : bool
            If True (default), series may include pages from multiple files.
            Currently applies to OME-TIFF only.
        multifile_close : bool
            If True (default), keep the handles of other files in multifile
            series closed. This is inefficient when few files refer to
            many pages. If False, the C runtime may run out of resources.
        fastij : bool
            If True (default), try to use only the metadata from the first page
            of ImageJ files. Significantly speeds up loading ImageJ
        movie : bool
            If True, assume that later pages differ from first page only by
            data offsets and byte_counts. Significantly increases speed and
            reduces memory usage when reading movies with thousands of pages.
            Python 3 only.
        kwargs : bool
            'is_ome': If False, disable processing of OME-XML metadata.

        for key, value in kwargs.items():
            if key[:3] == 'is_' and key[3:] in CONST.TIFF_FILE_ATTRS:
                if value is not None and not value:
                    setattr(self, key, bool(value))
                    if name == 'is_imagej':
                        fastij = False
                raise TypeError(
                    "got an unexpected keyword argument '%s'" % key)

        fh = FileHandle(arg, mode='rb', name=name, offset=offset, size=size)
        self._fh = fh
        self._multifile = bool(multifile)
        self._multifile_close = bool(multifile_close)
        self._files = {fh.name: self}  # cache of TiffFiles
                byteorder = {b'II': '<', b'MM': '>'}[fh.read(2)]
            except KeyError:
                raise ValueError("invalid TIFF file")
            sys_byteorder = {'big': '>', 'little': '<'}[sys.byteorder]
            self.is_native = byteorder == sys_byteorder

            version = struct.unpack(byteorder+'H', fh.read(2))[0]
            if version == 43:
                # BigTiff
                offset_size, zero = struct.unpack(byteorder+'HH', fh.read(4))
                if zero or offset_size != 8:
                    raise ValueError("invalid BigTIFF file")
                self.byteorder = byteorder
                self.offset_size = 8
                self.offset_format = byteorder+'Q'
                self.tagno_size = 8
                self.tagno_format = byteorder+'Q'
                self.tag_size = 20
                self.tag_format1 = byteorder+'HH'
                self.tag_format2 = byteorder+'Q8s'
            elif version == 42:
                self.byteorder = byteorder
                self.offset_size = 4
                self.offset_format = byteorder+'I'
                self.tagno_size = 2
                self.tagno_format = byteorder+'H'
                self.tag_size = 12
                self.tag_format1 = byteorder+'HH'
                self.tag_format2 = byteorder+'I4s'
                raise ValueError("not a TIFF file")

            # file handle is at offset to offset to first page
            self.pages = TiffPages(self, fastij=fastij)

            if movie:
                self.pages.useframes = True

            if self.is_lsm:
                # load all pages to fix offsets and byte_counts
                self.pages.useframes = True
                # second series: reduced/thumbnail pages
                self.pages.keyframe = 1
                keyframe = self.pages[1]
                for page in self.pages[1::2]:
                    page.keyframe = keyframe
                # first series: data
                self.pages.keyframe = 0
                keyframe = self.pages[0]
                for page in self.pages[::2]:
                    page.keyframe = keyframe
        except Exception:

    def filehandle(self):
        """Return file handle."""
        return self._fh

    def filename(self):
        """Return name of file handle."""
        return self._fh.name

    def fstat(self):
            return os.fstat(self._fh.fileno())
        except Exception:  # io.UnsupportedOperation
            return None

    def close(self):
        """Close open file handle(s)."""
        for tif in self._files.values():
        self._files = {}

    def asarray(self, key=None, series=None, memmap=False, tempdir=None):
        """Return image data from multiple TIFF pages as numpy array.

        By default the first image series is returned.

        key : int, slice, or sequence of page indices
            Defines which pages to return as array.
        series : int or TiffPageSeries
            Defines which series of pages to return as array.
        memmap : bool
            If True, return an read-only array stored in a binary file on disk
            if possible. The TIFF file is used if possible, else a temporary
            file is created.
        tempdir : str
            The directory where the memory-mapped file will be created.

        if not self.pages:
            return numpy.array([])
        if key is None and series is None:
            series = 0
        if series is not None:
                series = self.series[series]
            except (KeyError, TypeError):
            pages = series.pages
            pages = self.pages

        if key is None:
        elif isinstance(key, int):
            pages = [pages[key]]
        elif isinstance(key, slice):
            pages = pages[key]
        elif isinstance(key, collections.Iterable):
            pages = [pages[k] for k in key]
            raise TypeError("key must be an int, slice, or sequence")

        if not pages:
            raise ValueError("no pages selected")

        if self.is_nih:
            if pages[0].is_indexed:
                result = stack_pages(pages, colormapped=False, squeeze=False)
                result = apply_colormap(result, pages[0].color_map)
                result = stack_pages(pages, memmap=memmap, tempdir=tempdir,
                                     colormapped=False, squeeze=False)
        elif len(pages) == 1:
            result = pages[0].asarray(memmap=memmap)
        elif self.is_ome:
            if any(p is None for p in pages):
                # zero out missing pages
                firstpage = next(p for p in pages if p)
                nopage = numpy.zeros_like(
            if memmap:
                with tempfile.NamedTemporaryFile() as fh:
                    result = numpy.memmap(fh, series.dtype, shape=series.shape)
                    result = result.reshape(-1)
                result = numpy.empty(series.shape, series.dtype).reshape(-1)
            index = 0

            class KeepOpen(object):
                # keep Tiff files open between consecutive pages
                def __init__(self, parent, close):
                    self.master = parent
                    self.parent = parent
                    self._close = close

                def open(self, page):
                    if self._close and page and page.parent != self.parent:
                        if self.parent != self.master:
                        self.parent = page.parent

                def close(self):
                    if self._close and self.parent != self.master:

            keep = KeepOpen(self, self._multifile_close)
            for page in pages:
                if page:
                    a = page.asarray(memmap=False, colormapped=False,
                    a = nopage
                    result[index:index + a.size] = a.reshape(-1)
                except ValueError as e:
                    warnings.warn("ome-tiff: %s" % e)
                index += a.size
        elif key is None and series and series.offset:
            if memmap:
                result = self.filehandle.memmap_array(
                    series.dtype, series.shape, series.offset)
                result = self.filehandle.read_array(
                    series.dtype, product(series.shape))
            result = stack_pages(pages, memmap=memmap, tempdir=tempdir)

        if key is None:
                result.shape = series.shape
            except ValueError:
                    warnings.warn("failed to reshape %s to %s" % (
                        result.shape, series.shape))
                    # try series of expected shapes
                    result.shape = (-1,) + series.shape
                except ValueError:
                    # revert to generic shape
                    result.shape = (-1,) + pages[0].shape
        elif len(pages) == 1:
            result.shape = pages[0].shape
            result.shape = (-1,) + pages[0].shape
        return result

    def series(self):
        """Return associated pages as TiffPageSeries."""
        if not self.pages:
            return []

        series = []
        for name in 'ome imagej lsm fluoview nih shaped'.split():
            if getattr(self, 'is_' + name, False):
                series = getattr(self, '_%s_series' % name)()
        if not series:
            series = self._generic_series()

        # remove empty series, e.g. in MD Gel files
        series = [s for s in series if sum(s.shape) > 0]

        for i, s in enumerate(series):
            s.index = i
        return series

    def _generic_series(self):
        """Return image series in file."""
        self.pages.useframes = False

        shapes = []
        pages = {}
        for page in self.pages:
            if not page.shape:
            shape = page.shape + (
                page.axes, page.compression in CONST.TIFF_DECOMPESSORS)
            if shape in pages:
                pages[shape] = [page]
        series = []
        for s in shapes:
            shape = s[:-2]
            axes = s[-2]
            ps = pages[s]
            if len(ps) > 1:
                shape = (len(ps),) + shape
                axes = 'I' + axes
            series.append(TiffPageSeries(pages[s], shape, ps[0].dtype, axes))
        return series

    def _shaped_series(self):
        """Return image series in "shaped" file."""
        self.pages.useframes = True
        self.pages.keyframe = 0

        def append_series(series, pages, axes, shape, reshape):
            if not axes:
                shape = pages[0].shape
                axes = pages[0].axes
                if len(pages) > 1:
                    shape = (len(pages),) + shape
                    axes = 'I' + axes
                axes = reshape_axes(axes, shape, reshape)
                shape = reshape
            except ValueError as e:
            series.append(TiffPageSeries(pages, shape, pages[0].dtype, axes))

        keyframe = axes = shape = reshape = None
        series = []
        pages = []
        for page in self.pages:
            if not page.is_shaped:
                page.keyframe = keyframe
            if pages:
                append_series(series, pages, axes, shape, reshape)
            # new keyframe
            self.pages.keyframe = page.index
            keyframe = self.pages[page.index]
            pages = [keyframe]
            axes = None
            shape = None
            # read metadata
            metadata = json_description_dict(keyframe.is_shaped)
            reshape = metadata['shape']
            if 'axes' in metadata:
                axes = metadata['axes']
                if len(axes) == len(reshape):
                    shape = reshape
                    axes = ''
                    warnings.warn("axes do not match shape")
        if pages:
            append_series(series, pages, axes, shape, reshape)
        return series

    def _fluoview_series(self):
        """Return image series in FluoView file."""
        self.pages.useframes = True
        self.pages.keyframe = 0

        page = self.pages[0]
        mmhd = list(reversed(page.mm_header.dimensions))
        axes = ''.join(CONST.FLUOVIEW_DIMENSIONS.get(i[0].strip().upper(), 'Q')
                       for i in mmhd if i[1] > 1)
        shape = tuple(int(i[1]) for i in mmhd if i[1] > 1)
        return [TiffPageSeries(self.pages, shape, page.dtype, axes)]

    def _imagej_series(self):
        """Return image series in ImageJ file."""
        # ImageJ's dimension order is always TZCYXS
        # TODO: fix loading of color, composite or palette images
        self.pages.useframes = True
        self.pages.keyframe = 0
        shape = []
        axes = []
        page0 = self.pages[0]
        ij = page0.imagej_tags
        if 'frames' in ij:
        if 'slices' in ij:
        if 'channels' in ij and not (page0.is_rgb and not
                                     ij.get('hyperstack', False)):
        remain = ij.get('images', len(self.pages)) // (product(shape)
                                                       if shape else 1)
        if remain > 1:
        if page0.axes[0] == 'I':
            # contiguous multiple images
        elif page0.axes[:2] == 'SI':
            # color-mapped contiguous multiple images
            shape = page0.shape[0:1] + tuple(shape) + page0.shape[2:]
            axes = list(page0.axes[0]) + axes + list(page0.axes[2:])
        return [TiffPageSeries(self.pages, shape, page0.dtype, axes)]

    def _nih_series(self):
        """Return image series in NIH file."""
        self.pages.useframes = True
        self.pages.keyframe = 0
        page0 = self.pages[0]
        if len(self.pages) == 1:
            shape = page0.shape
            axes = page0.axes
            shape = (len(self.pages),) + page0.shape
            axes = 'I' + page0.axes
        return [TiffPageSeries(self.pages, shape, page0.dtype, axes)]

    def _ome_series(self):
        """Return image series in OME-TIFF file(s)."""
        from xml.etree import cElementTree as etree  # delayed import
        omexml = self.pages[0].tags['image_description'].value
            root = etree.fromstring(omexml)
        except etree.ParseError as e:
            # TODO: test this
            warnings.warn("ome-xml: %s" % e)
            omexml = omexml.decode('utf-8', 'ignore').encode('utf-8')
            root = etree.fromstring(omexml)

        self.pages.useframes = True
        self.pages.keyframe = 0

        uuid = root.attrib.get('UUID', None)
        self._files = {uuid: self}
        dirname = self._fh.dirname
        modulo = {}
        series = []
        for element in root:
            if element.tag.endswith('BinaryOnly'):
                warnings.warn("ome-xml: not an ome-tiff master file")
            if element.tag.endswith('StructuredAnnotations'):
                for annot in element:
                    if not annot.attrib.get('Namespace',
                    for value in annot:
                        for modul in value:
                            for along in modul:
                                if not along.tag[:-1].endswith('Along'):
                                axis = along.tag[-1]
                                newaxis = along.attrib.get('Type', 'other')
                                newaxis = CONST.AXES_LABELS[newaxis]
                                if 'Start' in along.attrib:
                                    step = float(along.attrib.get('Step', 1))
                                    start = float(along.attrib['Start'])
                                    stop = float(along.attrib['End']) + step
                                    labels = numpy.arange(start, stop, step)
                                    labels = [label.text for label in along
                                              if label.tag.endswith('Label')]
                                modulo[axis] = (newaxis, labels)

            if not element.tag.endswith('Image'):

            for pixels in element:
                if not pixels.tag.endswith('Pixels'):
                attr = pixels.attrib
                dtype = attr.get('PixelType', None)
                axes = ''.join(reversed(attr['DimensionOrder']))
                shape = list(int(attr['Size'+ax]) for ax in axes)
                size = product(shape[:-2])
                ifds = None
                spp = 1  # samples per pixel
                for data in pixels:
                    if data.tag.endswith('Channel'):
                        attr = data.attrib
                        if ifds is None:
                            spp = int(attr.get('SamplesPerPixel', spp))
                            ifds = [None] * (size // spp)
                        elif int(attr.get('SamplesPerPixel', 1)) != spp:
                            raise ValueError(
                                "Can't handle differing SamplesPerPixel")
                    if ifds is None:
                        ifds = [None] * (size // spp)
                    if not data.tag.endswith('TiffData'):
                    attr = data.attrib
                    ifd = int(attr.get('IFD', 0))
                    num = int(attr.get('NumPlanes', 1 if 'IFD' in attr else 0))
                    num = int(attr.get('PlaneCount', num))
                    idx = [int(attr.get('First'+ax, 0)) for ax in axes[:-2]]
                        idx = numpy.ravel_multi_index(idx, shape[:-2])
                    except ValueError:
                        # ImageJ produces invalid ome-xml when cropping
                        warnings.warn("ome-xml: invalid TiffData index")
                    for uuid in data:
                        if not uuid.tag.endswith('UUID'):
                        if uuid.text not in self._files:
                            if not self._multifile:
                                # abort reading multifile OME series
                                # and fall back to generic series
                                return []
                            fname = uuid.attrib['FileName']
                                tif = TiffFile(os.path.join(dirname, fname))
                            except (IOError, FileNotFoundError, ValueError):
                                    "ome-xml: failed to read '%s'" % fname)
                            self._files[uuid.text] = tif
                            if self._multifile_close:
                        pages = self._files[uuid.text].pages
                            for i in range(num if num else len(pages)):
                                ifds[idx + i] = pages[ifd + i]
                        except IndexError:
                            warnings.warn("ome-xml: index out of range")
                        # only process first uuid
                        pages = self.pages
                            for i in range(num if num else len(pages)):
                                ifds[idx + i] = pages[ifd + i]
                        except IndexError:
                            warnings.warn("ome-xml: index out of range")

                if all(i is None for i in ifds):
                    # skip images without data

                # set a keyframe on all ifds
                keyframe = None
                for i in ifds:
                    # try find a TiffPage
                    if i and i == i.keyframe:
                        keyframe = i
                if not keyframe:
                    # reload a TiffPage from file
                    for i, keyframe in enumerate(ifds):
                        if keyframe:
                            keyframe.parent.pages.keyframe = keyframe.index
                            keyframe = keyframe.parent.pages[keyframe.index]
                            ifds[i] = keyframe
                for i in ifds:
                    if i is not None:
                        i.keyframe = keyframe

                dtype = keyframe.dtype
                series.append(TiffPageSeries(ifds, shape, dtype, axes, self))
        for serie in series:
            shape = list(serie.shape)
            for axis, (newaxis, labels) in modulo.items():
                i = serie.axes.index(axis)
                size = len(labels)
                if shape[i] == size:
                    serie.axes = serie.axes.replace(axis, newaxis, 1)
                    shape[i] //= size
                    shape.insert(i+1, size)
                    serie.axes = serie.axes.replace(axis, axis+newaxis, 1)
            serie.shape = tuple(shape)
        # squeeze dimensions
        for serie in series:
            serie.shape, serie.axes = squeeze_axes(serie.shape, serie.axes)
        return series

    def _lsm_series(self):
        """Return main image series in LSM file. Skip thumbnails."""
        # all pages/frames are already loaded and corrected
        page0 = self.pages[0]
        lsmi = page0.cz_lsm_info
        axes = CONST.CZ_SCAN_TYPES[lsmi.scan_type]
        if page0.is_rgb:
            axes = axes.replace('C', '').replace('XY', 'XYC')
        if hasattr(lsmi, 'dimension_p') and lsmi.dimension_p > 1:
            axes += 'P'
        if hasattr(lsmi, 'dimension_m') and lsmi.dimension_m > 1:
            axes += 'M'
        axes = axes[::-1]
        shape = tuple(getattr(lsmi, CONST.CZ_DIMENSIONS[i]) for i in axes)

        pages = self.pages[::2]
        dtype = pages[0].dtype
        series = [TiffPageSeries(pages, shape, dtype, axes)]
        return series

    def _fix_lsm_strip_offsets(self):
        """Unwrap strip offsets for LSM files greater than 4 GB."""
        # each series and position require separate unwrapping (undocumented)
        for series in self.series:
            positions = 1
            for i in 0, 1:
                if series.axes[i] in 'PM':
                    positions *= series.shape[i]
            positions = len(series.pages) // positions
            for i, page in enumerate(series.pages):
                if not i % positions:
                    wrap = 0
                    previous_offset = 0
                data_offsets = []
                for current_offset in page.data_offsets:
                    if current_offset < previous_offset:
                        wrap += 2**32
                    data_offsets.append(current_offset + wrap)
                    previous_offset = current_offset
                page.data_offsets = tuple(data_offsets)

    def _fix_lsm_strip_byte_counts(self):
        """Set strip_byte_counts to size of compressed data.

        The strip_byte_counts tag in LSM files contains the number of bytes
        for the uncompressed data.

        pages = self.pages
        if not pages or not (pages[0].compression or pages[1].compression):
        strips = {}
        for page in pages:
            assert len(page.data_offsets) == len(page.data_byte_counts)
            for offset, count in zip(page.data_offsets, page.data_byte_counts):
                strips[offset] = count
        offsets = sorted(strips.keys())
        offsets.append(min(offsets[-1] + strips[offsets[-1]], self._fh.size))
        for i, offset in enumerate(offsets[:-1]):
            strips[offset] = min(strips[offset], offsets[i+1] - offset)
        if pages[0].compression:
            for page in pages[::2]:
                page.data_byte_counts = tuple(strips[offset]
                                              for offset in page.data_offsets)
        if pages[1].compression:
            for page in pages[1::2]:
                page.data_byte_counts = tuple(strips[offset]
                                              for offset in page.data_offsets)

    def __getattr__(self, name):
        """Return 'is_*' attributes from first page."""
        if name[3:] in CONST.TIFF_FILE_ATTRS:
            if not self.pages:
                return False
            value = bool(getattr(self.pages[0], name))
            setattr(self, name, value)
            return value
        raise AttributeError("'%s' object has no attribute '%s'" %
                             (self.__class__.__name__, name))

    def __len__(self):
        """Return number of pages in file."""
        return len(self.pages)

    def __getitem__(self, key):
        """Return specified page(s)."""
        return self.pages[key]

    def __iter__(self):
        """Return iterator over pages."""
        return iter(self.pages)

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_value, traceback):

    def __str__(self):
        """Return string containing information about file."""
        result = [
            "TiffFile '%s'" % self._fh.name,
            {'<': 'little endian', '>': 'big endian'}[self.byteorder]]
        if self.is_bigtiff:
        result.append('|'.join((attr for attr in CONST.TIFF_FILE_ATTRS
                                if getattr(self, 'is_' + attr))))
        if len(self.pages) > 1:
            result.append("%i pages" % len(self.pages))
        if len(self.series) > 1:
            result.append("%i series" % len(self.series))
        if len(self._files) > 1:
            result.append("%i files" % (len(self._files)))
        return "  ".join(result)

    def info(self, series=None, pages=None):
        """Return string with detailed information about file."""
        if series is None:
            series = self.series[:CONST.PRINT_MAX_LINES]
            series = [self.series[i] for i in sequence(series)]

        result = [str(self)]
        result.append("\n".join(str(s) for s in series))

        if pages is None:
            for s in series:
                result.append(next(p.info() for p in s.pages if p))
            if pages == 'all':
                pages = self.pages
                pages = [self.pages[i] for i in sequence(pages)]
            for p in pages:

        return '\n\n'.join(result)

    def micromanager_metadata(self):
        """Return MicroManager metadata not stored in TIFF tags."""
        if self.is_micromanager:
            return read_micromanager_metadata(self._fh)

    def scanimage_metadata(self):
        """Return ScanImage non-varying frame and ROI metadata as dict."""
        if self.is_scanimage:
                frame_data, roi_data = read_scanimage_metadata(self._fh)
                return frame_data
            except ValueError:

    def is_bigtiff(self):
        """File has BigTIFF format."""
        return self.offset_size != 4

    def is_mdgel(self):
        """File has MD Gel format."""
            return self.pages[1].is_mdgel
        except (IndexError, AttributeError):
            return False

class TiffPages(object):
    """Sequence of TIFF image file directories."""
    def __init__(self, parent, fastij=True):
        """Initialize instance from file. Read first TiffPage from file.

        The file position must be at an offset to an offset to a TiffPage.

        parent : TiffFile
            The TiffFile instance containing the pages.
        fastij : bool
            See TiffFile documentation.

        self.parent = parent
        self.offsets = []  # offsets to tiff pages in file
        self.pages = []  # cached tiff pages
        self.complete = False  # True if offsets to all pages were read
        self._tiffpage = TiffPage  # class for reading tiff pages
        self._keyframe = None
        self._cache = True

        # read offset to first page
        fh = parent.filehandle
        self._next_page_offset = fh.tell()
        offset = struct.unpack(parent.offset_format,

        if offset == 0:
            # warnings.warn("file contains no pages")
            self.complete = True
        if offset >= fh.size:
            warnings.warn("invalid page offset (%i)" % offset)
            self.complete = True

        # always read and cache first page
        page = TiffPage(parent, index=0)
        self._keyframe = page

        if (fastij and page.is_imagej and not page.is_ome
                and page.is_imagej_hyperstack):
            self.complete = True

    def cache(self):
        """Return if pages/frames are currenly being cached."""
        return self._cache

    def cache(self, value):
        """Enable or disable caching of pages/frames. Clear cache if False."""
        value = bool(value)
        if self._cache and not value:
        self._cache = value

    def useframes(self):
        """Return if currently using TiffFrame (True) or TiffPage (False)."""
        return self._tiffpage == TiffFrame

    def useframes(self, value):
        """Set to use TiffFrame (True) or TiffPage (False)."""
        self._tiffpage = TiffFrame if value else TiffPage

    def keyframe(self):
        """Return index of current keyframe."""
        return self._keyframe.index

    def keyframe(self, index):
        """Set current keyframe. Load TiffPage from file if necessary."""
        if self.complete or 0 <= index < len(self.offsets):
            page = self.pages[index]
            if isinstance(page, TiffPage):
                self._keyframe = page
                self.pages[index] = None  # remove existing frame
        # load TiffPage from file
        useframes = self.useframes
        self._tiffpage = TiffPage
        self._keyframe = self[index]
        self.useframes = useframes

    def next_page_offset(self):
        """Return offset where offset to a new page can be stored."""
        if not self.complete:
        return self._next_page_offset

    def load(self):
        """Read all remaining pages from file."""
        fh = self.parent.filehandle
        keyframe = self._keyframe
        if not self.complete:
        for i, (offset, page) in enumerate(zip(self.offsets, self.pages)):
            if page is None:
                page = self._tiffpage(self.parent, index=i, keyframe=keyframe)
                self.pages[i] = page

    def clear(self, fully=True):
        """Delete all but first page from cache. Set keyframe to first page."""
        if not self._cache or len(self.pages) < 1:
        self._keyframe = self.pages[0]
        if fully:
            # delete all but first
            self.pages = [self.pages[0]] + [None] * (len(self.pages)-1)
            # only delete TiffFrames
            for i, page in enumerate(self.pages):
                if page is not None and not isinstance(page, TiffFile):
                    self.pages[i] = None

    def _seek(self, index):
        """Seek file to offset of specified page."""
        fh = self.parent.filehandle

        if index is None:
            index = -1

        if self.complete or 0 <= index < len(self.offsets):

        offset_format = self.parent.offset_format
        offset_size = self.parent.offset_size
        tagno_format = self.parent.tagno_format
        tagno_size = self.parent.tagno_size
        tag_size = self.parent.tag_size
        unpack = struct.unpack

        while True:
            # read offsets to pages from file until index is reached
            offset = self.offsets[-1]

            # skip tags
                tagno = unpack(tagno_format, fh.read(tagno_size))[0]
                if tagno > 4096:
                    raise ValueError("suspicious number of tags")
            except Exception:
                warnings.warn("corrupted tag list at offset %i" % offset)
                del self.offsets[-1]
                self.complete = True
            self._next_page_offset = offset + tagno_size + tagno * tag_size

            # read offset to next page
            offset = unpack(offset_format, fh.read(offset_size))[0]
            if not offset:
                self.complete = True
            if offset >= fh.size:
                warnings.warn("invalid page offset (%i)" % offset)
                self.complete = True

            if index >= 0 and len(self.offsets) > index:

        if index >= len(self.offsets):
            raise IndexError()


    def __bool__(self):
        """Return True if file contains any pages."""
        return len(self.offsets) > 0

    def __len__(self):
        """Return number of pages in file."""
        if not self.complete:
        return len(self.offsets)

    def __getitem__(self, key):
        """Return specified page(s) from cache or file."""
        if isinstance(key, slice):
            start, stop, _ = key.indices(2**31)
            if not self.complete and max(stop, start) > len(self.offsets):
            return [self[i] for i in range(*key.indices(len(self.offsets)))]

        if self.complete and key >= len(self.offsets):
            raise IndexError()

            page = self.pages[key]
        except IndexError:
            page = None
        if page is not None:
            return page

        page = self._tiffpage(self.parent, index=key, keyframe=self._keyframe)
        if self._cache:
            self.pages[key] = page
        return page

    def __iter__(self):
        """Return iterator over all pages."""
        i = 0
        while True:
                yield self[i]
                i += 1
            except IndexError:

class TiffPage(object):
    """TIFF image file directory (IFD).

    index : int
        Index of page in file.
    dtype : str {TIFF_SAMPLE_DTYPES}
        Data type of image, color-mapped if applicable.
    shape : tuple
        Dimensions of the image array in TIFF page,
        color-mapped and with extra samples if applicable.
    axes : str
        Axes label codes:
        'X' width, 'Y' height, 'S' sample, 'I' image series|page|plane,
        'Z' depth, 'C' color|em-wavelength|channel, 'E' ex-wavelength|lambda,
        'T' time, 'R' region|tile, 'A' angle, 'P' phase, 'H' lifetime,
        'L' exposure, 'V' event, 'Q' unknown, '_' missing
    tags : dict
        Dictionary of tags in IFD. {tag.name: TiffTag}
    color_map : numpy.ndarray
        Color look up table, if exists.
    cz_lsm_scan_info: dict
        LSM scan info attributes, if exists.
    imagej_tags: dict
        Consolidated ImageJ description and metadata tags, if exists.
    uic_tags: dict
        Consolidated MetaMorph STK/UIC tags, if exists.

    All attributes are read-only.

    The internal, normalized '_shape' attribute is 6 dimensional:

    0 : number planes/images  (stk, ij).
    1 : planar samples_per_pixel.
    2 : image_depth Z  (sgi).
    3 : image_length Y.
    4 : image_width X.
    5 : contig samples_per_pixel.

    def __init__(self, parent, index, keyframe=None):
        """Initialize instance from file.

        The file handle must be at offset to a valid IFD.

        self.parent = parent
        self.index = index

        self.shape = self._shape = ()
        self.dtype = self._dtype = None
        self.axes = ""
        self.offset = None  # offset to this IDF
        self.tags = {}


    def _fromfile(self):
        """Read TIFF IFD structure and its tags from file."""
        parent = self.parent
        fh = parent.filehandle

        self.offset = fh.tell()  # offset to this IDF

        # read standard tags
            tagno = struct.unpack(parent.tagno_format,
            if tagno > 4096:
                raise ValueError("suspicious number of tags")
        except Exception:
            raise ValueError("corrupted tag list at offset %i" % self.offset)

        tags = self.tags
        for _ in range(tagno):
                tag = TiffTag(self.parent)
            except TiffTag.Error as e:
            if tag.name not in tags:
                tags[tag.name] = tag
                # some files contain multiple tags with same code
                # e.g. MicroManager files contain two image_description tags
                i = 1
                while True:
                    name = "%s_%i" % (tag.name, i)
                    if name not in tags:
                        tags[name] = tag

        if self.is_andor:
            # consolidate Andor tags; remove them from self.tags

        if self.is_lsm or (self.index and self.parent.is_lsm):
            # correct non standard LSM bitspersample tags

        if self.is_lsm:
            # read LSM info subrecords
            for name, reader in CONST.CZ_LSM_INFO_READERS.items():
                    offset = self.cz_lsm_info['offset_'+name]
                except KeyError:
                if offset < 8:
                    # older LSM revision
                    setattr(self, 'cz_lsm_'+name, reader(fh))
                except ValueError:
        elif self.is_stk and 'uic1tag' in tags and not tags['uic1tag'].value:
            # read uic1tag now that plane count is known
            uic1tag = tags['uic1tag']
            tags['uic1tag'].value = read_uic1tag(fh, parent.byteorder,
                                                 uic1tag.dtype, uic1tag.count,

    def _process_tags(self):
        """Validate standard tags and initialize attributes.

        Raise ValueError if tag values are not supported.

        tags = self.tags

        if 'bits_per_sample' in tags:
            tag = tags['bits_per_sample']
            if tag.count == 1:
                self.bits_per_sample = tag.value
                # LSM might list more items than samples_per_pixel
                value = tag.value[:self.samples_per_pixel]
                if any((v-value[0] for v in value)):
                    self.bits_per_sample = value
                    self.bits_per_sample = value[0]

        if 'sample_format' in tags:
            tag = tags['sample_format']
            if tag.count == 1:
                self.sample_format = CONST.TIFF_SAMPLE_FORMATS[tag.value]
                value = tag.value[:self.samples_per_pixel]
                if any((v-value[0] for v in value)):
                    self.sample_format = [CONST.TIFF_SAMPLE_FORMATS[v]
                                          for v in value]
                    self.sample_format = CONST.TIFF_SAMPLE_FORMATS[value[0]]

        if 'photometric' not in tags:
            self.photometric = None

        if 'image_length' in tags:
            if 'rows_per_strip' not in tags:
                self.rows_per_strip = self.image_length
            self.strips_per_image = int(math.floor(
                float(self.image_length + self.rows_per_strip - 1) /
            self.strips_per_image = 0

        key = (self.sample_format, self.bits_per_sample)
        self.dtype = self._dtype = CONST.TIFF_SAMPLE_DTYPES.get(key, None)

        if 'image_length' not in self.tags or 'image_width' not in self.tags:
            # some GEL file pages are missing image data
            self.image_length = 0
            self.image_width = 0
            self.image_depth = 0
            self.strip_offsets = (0,)
            self._shape = ()
            self.shape = ()
            self.axes = ''

        if self.is_vista or (self.index and self.parent.is_vista):
            # ISS Vista writes wrong image_depth tag
            self.image_depth = 1

        if self.is_indexed:
            self.dtype = self.tags['color_map'].dtype[1]
            self.color_map = numpy.array(self.color_map, self.dtype)
            dmax = self.color_map.max()
            if dmax < 256:
                self.dtype = numpy.uint8
                self.color_map = self.color_map.astype(self.dtype)
            # else:
            #     self.dtype = numpy.uint8
            #     self.color_map >>= 8
            #     self.color_map = self.color_map.astype(self.dtype)
            # TODO: support other photometric modes than RGB
            self.color_map.shape = (3, -1)

        # determine shape of data
        image_length = self.image_length
        image_width = self.image_width
        image_depth = self.image_depth
        samples_per_pixel = self.samples_per_pixel

        if self.is_stk:
            assert self.image_depth == 1
            planes = self.tags['uic2tag'].count
            if self.is_contig:
                self._shape = (planes, 1, 1, image_length, image_width,
                if samples_per_pixel == 1:
                    self.shape = (planes, image_length, image_width)
                    self.axes = 'YX'
                    self.shape = (planes, image_length, image_width,
                    self.axes = 'YXS'
                self._shape = (planes, samples_per_pixel, 1, image_length,
                               image_width, 1)
                if samples_per_pixel == 1:
                    self.shape = (planes, image_length, image_width)
                    self.axes = 'YX'
                    self.shape = (planes, samples_per_pixel, image_length,
                    self.axes = 'SYX'
            # detect type of series
            if planes == 1:
                self.shape = self.shape[1:]
            elif numpy.all(self.uic2tag['z_distance'] != 0):
                self.axes = 'Z' + self.axes
            elif numpy.all(numpy.diff(self.uic2tag['time_created']) != 0):
                self.axes = 'T' + self.axes
                self.axes = 'I' + self.axes
            # DISABLED
            if self.is_indexed:
                assert False, "color mapping disabled for stk"
                if self.color_map.shape[1] >= 2**self.bits_per_sample:
                    if image_depth == 1:
                        self.shape = (planes, image_length, image_width,
                        self.shape = (planes, image_depth, image_length,
                                      image_width, self.color_map.shape[0])
                    self.axes = self.axes + 'S'
                    warnings.warn("palette cannot be applied")
                    self.is_indexed = False
        elif self.is_indexed:
            samples = 1
            if 'extra_samples' in self.tags:
                samples += self.tags['extra_samples'].count
            if self.is_contig:
                self._shape = (1, 1, image_depth, image_length, image_width,
                self._shape = (1, samples, image_depth, image_length,
                               image_width, 1)
            if self.color_map.shape[1] >= 2**self.bits_per_sample:
                if image_depth == 1:
                    self.shape = (image_length, image_width,
                    self.axes = 'YXS'
                    self.shape = (image_depth, image_length, image_width,
                    self.axes = 'ZYXS'
                warnings.warn("palette cannot be applied")
                self.is_indexed = False
                if image_depth == 1:
                    self.shape = (image_length, image_width)
                    self.axes = 'YX'
                    self.shape = (image_depth, image_length, image_width)
                    self.axes = 'ZYX'
        elif self.is_rgb or samples_per_pixel > 1:
            if self.is_contig:
                self._shape = (1, 1, image_depth, image_length, image_width,
                if image_depth == 1:
                    self.shape = (image_length, image_width, samples_per_pixel)
                    self.axes = 'YXS'
                    self.shape = (image_depth, image_length, image_width,
                    self.axes = 'ZYXS'
                self._shape = (1, samples_per_pixel, image_depth,
                               image_length, image_width, 1)
                if image_depth == 1:
                    self.shape = (samples_per_pixel, image_length, image_width)
                    self.axes = 'SYX'
                    self.shape = (samples_per_pixel, image_depth,
                                  image_length, image_width)
                    self.axes = 'SZYX'
            if False and self.is_rgb and 'extra_samples' in self.tags:
                # DISABLED: only use RGB and first alpha channel if exists
                extra_samples = self.extra_samples
                if self.tags['extra_samples'].count == 1:
                    extra_samples = (extra_samples,)
                for exs in extra_samples:
                    if exs in ('unassalpha', 'assocalpha', 'unspecified'):
                        if self.is_contig:
                            self.shape = self.shape[:-1] + (4,)
                            self.shape = (4,) + self.shape[1:]
            self._shape = (1, 1, image_depth, image_length, image_width, 1)
            if image_depth == 1:
                self.shape = (image_length, image_width)
                self.axes = 'YX'
                self.shape = (image_depth, image_length, image_width)
                self.axes = 'ZYX'

        # data_offsets and data_byte_counts
        self.data_byte_counts = None
        if 'tile_offsets' in self.tags:
            self.data_offsets = self.tile_offsets
            self.data_offsets = self.strip_offsets
        if 'tile_byte_counts' in tags:
            self.data_byte_counts = self.tile_byte_counts
        elif 'strip_byte_counts' in tags:
            self.data_byte_counts = self.strip_byte_counts
        if self.data_byte_counts is None:
            if not self.compression:
                self.data_byte_counts = (
                    product(self.shape) * (self.bits_per_sample // 8),)
                raise ValueError("byte_counts not found")

        assert len(self.shape) == len(self.axes)

    def is_imagej_hyperstack(self):
        """Return if ImageJ data are contiguous and adjust page attributes.

        Patch 'data_offsets' and 'data_byte_counts' tags to span the
        complete contiguous data.

        ImageJ stores all image metadata in the first page and image data is
        stored contiguously before the second page, if any. No need to
        read other pages.

        parent = self.parent
        if not self.is_imagej or not self.is_contiguous:
        images = self.imagej_tags.get('images', 0)
        if images <= 1:
        offset, count = self.is_contiguous
        shape = self.shape
        if self.is_indexed:
            shape = shape[:-1]

        fh = parent.filehandle
        if (count != product(shape) * self.bits_per_sample // 8 or
                offset + count*images > fh.size):
            self.is_imagej = False
            warnings.warn("invalid ImageJ metadata or corrupted file")

        # check that next page is stored after data
        pos = fh.tell()
        nextpage = struct.unpack(parent.offset_format,
        if nextpage and offset + count*images > nextpage:

        # patch metadata
        self.data_offsets = (offset,)
        self.data_byte_counts = (count * images,)
        self.shape = (images,) + self.shape
        self._shape = (images,) + self._shape[1:]
        self.axes = 'I' + self.axes
        return True

    def asarray(self, squeeze=True, colormapped=True, rgbonly=False,
                scale_mdgel=False, memmap=False, reopen=True,
                maxsize=64*2**30, validate=True):
        """Read image data from file and return as numpy array.

        Raise ValueError if format is unsupported.
        If any of 'squeeze', 'colormapped', or 'rgbonly' are not the default,
        the shape of the returned array might be different from the page shape.

        squeeze : bool
            If True, all length-1 dimensions (except X and Y) are
            squeezed out from result.
        colormapped : bool
            If True, color mapping is applied for palette-indexed images.
        rgbonly : bool
            If True, return RGB(A) image without additional extra samples.
        memmap : bool
            If True, use numpy.memmap to read arrays from file if possible.
            For use on 64-bit systems and files with few huge contiguous data.
        reopen : bool
            If True and the parent file handle is closed, the file is
            temporarily re-opened (and closed if no exception occurs).
        scale_mdgel : bool
            If True, MD Gel data will be scaled according to the private
            metadata in the second TIFF page. The dtype will be float32.
        maxsize: int or None
            Maximum size of data before a ValueError is raised.
            Can be used to catch DOS. Default: 64 GB.

        self_ = self
        self = self.keyframe  # self or keyframe

        if not self._shape:

        tags = self.tags

        if validate:
            if maxsize and product(self._shape) > maxsize:
                raise ValueError("data is too large %s" % str(self._shape))

            if self.dtype is None:
                raise ValueError("data type not supported: %s%i" % (
                    self.sample_format, self.bits_per_sample))
            if self.compression not in CONST.TIFF_DECOMPESSORS:
                raise ValueError("cannot decompress %s" % self.compression)
            if 'sample_format' in tags:
                tag = tags['sample_format']
                if tag.count != 1 and any((i-tag.value[0] for i in tag.value)):
                    raise ValueError(
                        "sample formats do not match %s" % tag.value)

            if self.is_chroma_subsampled:
                # TODO: implement chroma subsampling
                raise NotImplementedError("chroma subsampling not supported")

        fh = self_.parent.filehandle
        closed = fh.closed
        if closed:
            if reopen:
                raise IOError("file handle is closed")

        dtype = self._dtype
        shape = self._shape
        image_width = self.image_width
        image_length = self.image_length
        image_depth = self.image_depth
        typecode = self.parent.byteorder + dtype
        bits_per_sample = self.bits_per_sample
        lsb2msb = self.fill_order == 'lsb2msb'

        offsets, byte_counts = self_.offsets_byte_counts

        if self.is_tiled:
            tile_width = self.tile_width
            tile_length = self.tile_length
            tile_depth = self.tile_depth if 'tile_depth' in tags else 1
            tw = (image_width + tile_width - 1) // tile_width
            tl = (image_length + tile_length - 1) // tile_length
            td = (image_depth + tile_depth - 1) // tile_depth
            shape = (shape[0], shape[1],
                     td*tile_depth, tl*tile_length, tw*tile_width, shape[-1])
            tile_shape = (tile_depth, tile_length, tile_width, shape[-1])
            runlen = tile_width
            runlen = image_width

        if memmap and self.is_memmappable(rgbonly, colormapped):
            result = fh.memmap_array(typecode, shape, offset=offsets[0])
        elif self.is_contiguous:
            result = fh.read_array(typecode, product(shape))
            result = result.astype('=' + dtype)
            if lsb2msb:
            if self.is_contig:
                runlen *= self.samples_per_pixel
            if bits_per_sample in (8, 16, 32, 64, 128):
                if (bits_per_sample * runlen) % 8:
                    raise ValueError("data and sample size mismatch")

                def unpack(x, typecode=typecode):
                    if self.predictor == 'float':
                        # the floating point horizontal differencing decoder
                        # needs the raw byte order
                        typecode = dtype
                        return numpy.fromstring(x, typecode)
                    except ValueError as e:
                        # strips may be missing EOI
                        warnings.warn("unpack: %s" % e)
                        xlen = ((len(x) // (bits_per_sample // 8)) *
                                (bits_per_sample // 8))
                        return numpy.fromstring(x[:xlen], typecode)

            elif isinstance(bits_per_sample, tuple):
                def unpack(x):
                    return unpack_rgb(x, typecode, bits_per_sample)
                def unpack(x):
                    return unpack_ints(x, typecode, bits_per_sample, runlen)

            decompress = CONST.TIFF_DECOMPESSORS[self.compression]
            if self.compression == 'jpeg':
                table = self.jpeg_tables if 'jpeg_tables' in tags else b''

                def decompress(x):
                    return decode_jpeg(x, table, self.photometric)

            if self.is_tiled:
                result = numpy.empty(shape, dtype)
                tw, tl, td, pl = 0, 0, 0, 0
                for offset, bytecount in zip(offsets, byte_counts):
                    tile = fh.read(bytecount)
                    if lsb2msb:
                        tile = reverse_bitorder(tile)
                    tile = decompress(tile)
                    tile = unpack(tile)
                        tile.shape = tile_shape
                    except ValueError:
                        # incomplete tiles; see gdal issue #1179
                        warnings.warn("invalid tile data")
                        t = numpy.zeros(tile_shape, dtype).reshape(-1)
                        s = min(tile.size, t.size)
                        t[:s] = tile[:s]
                        tile = t.reshape(tile_shape)
                    if self.predictor == 'horizontal':
                        numpy.cumsum(tile, axis=-2, dtype=dtype, out=tile)
                    elif self.predictor == 'float':
                        raise NotImplementedError()
                    result[0, pl, td:td+tile_depth,
                           tl:tl+tile_length, tw:tw+tile_width, :] = tile
                    del tile
                    tw += tile_width
                    if tw >= shape[4]:
                        tw, tl = 0, tl + tile_length
                        if tl >= shape[3]:
                            tl, td = 0, td + tile_depth
                            if td >= shape[2]:
                                td, pl = 0, pl + 1
                result = result[...,
                                :image_depth, :image_length, :image_width, :]
                strip_size = self.rows_per_strip * self.image_width
                if self.planar_configuration == 'contig':
                    strip_size *= self.samples_per_pixel
                result = numpy.empty(shape, dtype).reshape(-1)
                index = 0
                for offset, bytecount in zip(offsets, byte_counts):
                    strip = fh.read(bytecount)
                    if lsb2msb:
                        strip = reverse_bitorder(strip)
                    strip = decompress(strip)
                    strip = unpack(strip)
                    size = min(result.size, strip.size, strip_size,
                               result.size - index)
                    result[index:index+size] = strip[:size]
                    del strip
                    index += size

        result.shape = self._shape

        if self.predictor and not (self.is_tiled and not self.is_contiguous):
            if self.parent.is_lsm and not self.compression:
                pass  # work around bug in LSM510 software
            elif self.predictor == 'horizontal':
                numpy.cumsum(result, axis=-2, dtype=dtype, out=result)
            elif self.predictor == 'float':
                result = decode_floats(result)
        if colormapped and self.is_indexed:
            if self.color_map.shape[1] >= 2**bits_per_sample:
                # FluoView and LSM might fail here
                result = apply_colormap(result[:, 0:1, :, :, :, 0:1],
        elif rgbonly and self.is_rgb and 'extra_samples' in tags:
            # return only RGB and first alpha channel if exists
            extra_samples = self.extra_samples
            if tags['extra_samples'].count == 1:
                extra_samples = (extra_samples,)
            for i, exs in enumerate(extra_samples):
                if exs in ('unassalpha', 'assocalpha', 'unspecified'):
                    if self.is_contig:
                        result = result[..., [0, 1, 2, 3+i]]
                        result = result[:, [0, 1, 2, 3+i]]
                if self.is_contig:
                    result = result[..., :3]
                    result = result[:, :3]

        if squeeze:
                result.shape = self.shape
            except ValueError:
                warnings.warn("failed to reshape from %s to %s" % (
                    str(result.shape), str(self.shape)))

        if scale_mdgel and self.parent.is_mdgel:
            # MD Gel stores private metadata in the second page
            page1 = self.parent.pages[1]
            if page1.md_file_tag in (2, 128):
                scale = page1.md_scale_pixel
                scale = scale[0] / scale[1]  # rational
                result = result.astype('float32')
                if page1.md_file_tag == 2:
                    result **= 2  # squary root data format
                result *= scale

        if closed:
            # TODO: file should remain open if an exception occurred above
        return result

    def aspage(self):
        return self

    def keyframe(self):
        return self

    def keyframe(self, index):

    def offsets_byte_counts(self):
        """Return simplified offsets and byte_counts."""
        if self.is_contiguous:
            offset, byte_count = self.is_contiguous
            return [offset], [byte_count]
        return clean_offsets_counts(self.data_offsets, self.data_byte_counts)

    def is_contiguous(self):
        """Return offset and size of contiguous data, else None.

        Excludes prediction, fill_order, and colormapping.

        if self.compression or self.bits_per_sample not in (8, 16, 32, 64):
        if self.is_tiled:
            if (self.image_width != self.tile_width or
                    self.image_length % self.tile_length or
                    self.tile_width % 16 or self.tile_length % 16):
            if ('image_depth' in self.tags and 'tile_depth' in self.tags and
                    (self.image_length != self.tile_length or
                     self.image_depth % self.tile_depth)):

        offsets = self.data_offsets
        byte_counts = self.data_byte_counts
        if len(offsets) == 1:
            return offsets[0], byte_counts[0]
        if self.is_stk or all(offsets[i] + byte_counts[i] == offsets[i+1] or
                              byte_counts[i+1] == 0  # no data/ignore offset
                              for i in range(len(offsets)-1)):
            return offsets[0], sum(byte_counts)

    def is_memmappable(self, rgbonly, colormapped):
        """Return if page's image data in file can be memory-mapped."""
        return (self.parent.filehandle.is_file and
                self.is_contiguous and
                (self.bits_per_sample == 8 or self.parent.is_native) and
                self.fill_order == 'msb2lsb' and
                not self.predictor and
                not self.is_chroma_subsampled and
                not (rgbonly and 'extra_samples' in self.tags) and
                not (colormapped and self.is_indexed))

    def __getattr__(self, name):
        """Return and validate tag value."""
        if name in self.tags:
            tag = self.tags[name]
            if tag.code in CONST.TIFF_TAGS:
                (name, _, _, _, validate) = CONST.TIFF_TAGS[tag.code]
                if validate:
                        if tag.count == 1:
                            value = validate[tag.value]
                            value = tuple(validate[val] for val in tag.value)
                    except KeyError:
                        raise ValueError(
                            "%s.value (%s) not supported" % (name, tag.value))
                    value = tag.value
                value = tag.value
            setattr(self, name, value)
            return value
        if name in CONST.TIFF_TAGS_DEFAULTS:
            value = CONST.TIFF_TAGS_DEFAULTS[name]
            setattr(self, name, value)
            return value
        raise AttributeError("'%s' object has no attribute '%s'" %
                             (self.__class__.__name__, name))

    def __str__(self):
        """Return string containing information about page."""
        s = '  '.join(s for s in (
            'x'.join(str(i) for i in self.shape),
            '%s bit' % str(self.bits_per_sample),
            self.photometric if 'photometric' in self.tags else '',
            self.compression if self.compression else 'raw',
            '|'.join(t for t in CONST.TIFF_PAGE_ATTRS
                     if getattr(self, 'is_' + t)),
            '|'.join(t for t in CONST.TIFF_FILE_ATTRS
                     if getattr(self, 'is_' + t))
            ) if s)
        return "TiffPage %i  %s" % (self.index, s)

    def info(self, verbose=True):
        """Return string with detailed information about page."""
        result = [str(self)]
        tags = self.tags
                      for t in sorted(tags.values(), key=lambda x: x.code)))
        if verbose:
            for name in ('image_description', 'software', 'artist'):
                if name not in tags:
                value = tags[name].value
                if len(value) < CONST.PRINT_LINE_WIDTH - len(name) - 20:
                    result.append('%s\n%s' % (name.upper(), pprint(value)))
                except Exception:
            if self.is_indexed:
                result.append('Color Map: %s, %s' % (self.color_map.shape,
            for attr in ('cz_lsm_info', 'cz_lsm_scan_info', 'uic_tags',
                         'mm_header', 'imagej_tags', 'micromanager_metadata',
                         'nih_image_header', 'tvips_metadata', 'sfeg_metadata',
                         'metaseries_metadata', 'helios_metadata',
                         'sem_metadata', 'andor_tags'):
                if hasattr(self, attr) and getattr(self, attr, False):
                                             pprint(getattr(self, attr)))))

        return '\n\n'.join(result)

    def ndim(self):
        """Return number of array dimensions."""
        return len(self.shape)

    def size(self):
        """Return number of elements in array."""
        return int(product(self.shape))

    def uic_tags(self):
        """Return consolidated UIC tags."""
        if not self.is_stk:
            raise AttributeError("'%s' object has no attribute 'uic_tags'" %
        tags = self.tags
        result = {}
        result['number_planes'] = tags['uic2tag'].count
        if 'image_description' in tags:
            result['plane_descriptions'] = self.image_description.split(b'\0')
            # result['plane_descriptions'] = stk_description_dict(
            #    self.image_description)
        if 'uic1tag' in tags:
        if 'uic3tag' in tags:
            result.update(tags['uic3tag'].value)  # wavelengths
        if 'uic4tag' in tags:
            result.update(tags['uic4tag'].value)  # override uic1 tags
        uic2tag = tags['uic2tag'].value
        result['z_distance'] = uic2tag['z_distance']
        result['time_created'] = uic2tag['time_created']
        result['time_modified'] = uic2tag['time_modified']
            result['datetime_created'] = numpy.array(
                [julian_datetime(*dt) for dt in
                 zip(uic2tag['date_created'], uic2tag['time_created'])],
            result['datetime_modified'] = numpy.array(
                [julian_datetime(*dt) for dt in
                 zip(uic2tag['date_modified'], uic2tag['time_modified'])],
        except ValueError as e:
            warnings.warn("uic_tags: %s" % e)
        return result

    def andor_tags(self):
        """Return consolidated Andor tags. Remove Andor tags from self.tags."""
        if not self.is_andor:
            raise AttributeError("'%s' object has no attribute 'andor_tags'" %
        tags = self.tags
        result = {'id': bytes2str(tags['andor'].value)}
        for tag in list(self.tags.values()):
            code = tag.code
            if 4864 < code < 5031:
                key = CONST.ANDOR_TAGS.get(code, code)
                value = tag.value
                if isinstance(value, bytes):
                    value = bytes2str(value)
                result[key] = value
                del tags[tag.name]
        return result

    def metaseries_metadata(self):
        """Return MetaSeries tags from image_description."""
        if not self.is_metaseries:
            raise AttributeError(
                "'%s' object has no attribute 'metaseries_metadata'" %
        return metaseries_description_dict(self.image_description)

    def imagej_tags(self):
        """Return consolidated ImageJ metadata as dict."""
        if not self.is_imagej:
            raise AttributeError("'%s' object has no attribute 'imagej_tags'" %
        result = imagej_description_dict(self.is_imagej)
        if 'imagej_metadata' in self.tags:
            except Exception as e:
        return result

    def is_indexed(self):
        """Page contains indexed, palette-colored image.

        Disable color-mapping for OME, LSM, STK, and ImageJ hyperstacks.

        if (self.is_stk
                or self.is_lsm or (self.index and self.parent.is_lsm)
                or self.is_ome or (self.index and self.parent.is_ome)):
            return False
        if self.is_imagej:
            if b'mode' in self.is_imagej:
                return False
        elif (self.index and self.parent.is_imagej):
            return self.pages[0].is_indexed
        return ('photometric' in self.tags and
                self.tags['photometric'].value == 3)

    def is_rgb(self):
        """Page contains RGB image."""
        return ('photometric' in self.tags and
                self.tags['photometric'].value == 2)

    def is_contig(self):
        """Page contains contiguous image."""
        if 'planar_configuration' in self.tags:
            return self.tags['planar_configuration'].value == 1
        return True

    def is_tiled(self):
        """Page contains tiled image."""
        return 'tile_width' in self.tags

    def is_reduced(self):
        """Page is reduced image of another image."""
        return ('new_subfile_type' in self.tags and
                self.tags['new_subfile_type'].value & 1)

    def is_chroma_subsampled(self):
        """Page contains chroma subsampled image."""
        return ('ycbcr_subsampling' in self.tags and
                self.tags['ycbcr_subsampling'].value != (1, 1))

    def is_imagej(self):
        """Return ImageJ description if exists, else None."""
        if 'image_description' in self.tags:
            description = self.tags['image_description'].value
            if description.startswith(b'ImageJ='):
                return description
        if 'image_description_1' in self.tags:
            # Micromanager
            description = self.tags['image_description_1'].value
            if description.startswith(b'ImageJ='):
                return description

    def is_shaped(self):
        """Return description containing array shape if exists, else None."""
        if 'image_description' in self.tags:
            descr = self.tags['image_description'].value
            if b'"shape": [' in descr or descr.startswith(b'shape='):
                return descr
            if 'image_description_1' in self.tags:
                descr = self.tags['image_description_1'].value
                if b'"shape": [' in descr or descr.startswith(b'shape='):
                    return descr

    def is_mdgel(self):
        """Page contains md_file_tag tag."""
        return 'md_file_tag' in self.tags

    def is_mediacy(self):
        """Page contains Media Cybernetics Id tag."""
        return ('mc_id' in self.tags and
                self.tags['mc_id'].value.startswith(b'MC TIFF'))

    def is_stk(self):
        """Page contains UIC2Tag tag."""
        return 'uic2tag' in self.tags

    def is_lsm(self):
        """Page contains LSM CZ_LSM_INFO tag."""
        return 'cz_lsm_info' in self.tags

    def is_fluoview(self):
        """Page contains FluoView MM_STAMP tag."""
        return 'mm_stamp' in self.tags

    def is_nih(self):
        """Page contains NIH image header."""
        return 'nih_image_header' in self.tags

    def is_sgi(self):
        """Page contains SGI image and tile depth tags."""
        return 'image_depth' in self.tags and 'tile_depth' in self.tags

    def is_vista(self):
        """Software tag is 'ISS Vista'."""
        return ('software' in self.tags and
                self.tags['software'].value == b'ISS Vista')

    def is_metaseries(self):
        """Page contains MDS MetaSeries metadata in image_description tag."""
        if (self.index > 1
                or 'image_description' not in self.tags
                or 'software' not in self.tags
                or self.tags['software'].value != b'MetaSeries'):
            return False
        d = self.tags['image_description'].value.strip()
        return d.startswith(b'<MetaData>') and d.endswith(b'</MetaData>')

    def is_ome(self):
        """Page contains OME-XML in image_description tag."""
        if self.index > 1 or 'image_description' not in self.tags:
            return False
        d = self.tags['image_description'].value.strip()
        return d.startswith(b'<?xml version=') and d.endswith(b'</OME>')

    def is_scn(self):
        """Page contains Leica SCN XML in image_description tag."""
        if self.index > 1 or 'image_description' not in self.tags:
            return False
        d = self.tags['image_description'].value.strip()
        return d.startswith(b'<?xml version=') and d.endswith(b'</scn>')

    def is_micromanager(self):
        """Page contains Micro-Manager metadata."""
        return 'micromanager_metadata' in self.tags

    def is_andor(self):
        """Page contains Andor Technology tags."""
        return 'andor' in self.tags

    def is_tvips(self):
        """Page contains TVIPS metadata."""
        return 'tvips_metadata' in self.tags

    def is_fei(self):
        """Page contains SFEG or HELIOS metadata."""
        return 'sfeg_metadata' in self.tags or 'helios_metadata' in self.tags

    def is_sem(self):
        """Page contains Zeiss SEM metadata."""
        return 'sem_metadata' in self.tags

    def is_svs(self):
        """Page contains Aperio metadata."""
        return ('image_description' in self.tags and
                self.image_description.startswith(b'Aperio Image Library'))

    def is_scanimage(self):
        """Page contains ScanImage metadata."""
        return (('image_description' in self.tags and
                 self.image_description.startswith(b'state.config')) or
                ('software' in self.tags and

class TiffFrame(object):
    """Lightweight TIFF image file directory (IFD).

    Only a limited number of tag values are read from file, e.g. strip_offsets,
    and strip_byte_counts. Other tag values are assumed to be identical with a
    specified TiffPage instance, the keyframe.

    This is intended to reduce resource usage and speed up reading data from
    file, not for introspection of metadata.

    Not compatible with Python 2.

    tags = {}  # TiffFrame instances have no tags

    def __init__(self, parent, index, keyframe, tags=None):
        """Read specified tags from file.

        The file handle must be at offset to a valid IFD.

        self.keyframe = keyframe
        self.parent = parent
        self.index = index

        unpack = struct.unpack
        fh = parent.filehandle
        self.offset = fh.tell()
            tagno = unpack(parent.tagno_format, fh.read(parent.tagno_size))[0]
            if tagno > 4096:
                raise ValueError("suspicious number of tags")
        except Exception:
            raise ValueError("corrupted page list at offset %i" % self.offset)

        if tags is None:
            tags = CONST.TIFF_FRAME_TAGS

        tag_size = parent.tag_size
        code_format = parent.tag_format1[:2]

        for _ in range(tagno):
            data = fh.read(tag_size)
            code = unpack(code_format, data[:2])[0]
            if code not in tags:
            fh.seek(-tag_size, 1)
                tag = TiffTag(parent)
            except TiffTag.Error as e:
            if code == 273 or code == 324:
                setattr(self, 'data_offsets', tag.value)
            elif code == 279 or code == 325:
                setattr(self, 'data_byte_counts', tag.value)
                tagname = tag.name
                tagvalue = tag.value
                if tagname not in self.__dict__:
                    setattr(self, tagname, tagvalue)
                    # some files contain multiple tags with same code
                    i = 1
                    while True:
                        name = "%s_%i" % (tagname, i)
                        if name not in self.__dict__:
                            setattr(self, name, tagvalue)

    def aspage(self):
        """Return TiffPage from file."""
        return TiffPage(self.parent, index=self.index, keyframe=None)

    def asarray(self, *args, **kwargs):
        """Read image data from file and return as numpy array."""
        # TODO: fix TypeError on Python 2
        #   "TypeError: unbound method asarray() must be called with TiffPage
        #   instance as first argument (got TiffFrame instance instead)"
        kwargs['validate'] = False
        return TiffPage.asarray(self, *args, **kwargs)

    def offsets_byte_counts(self):
        """Return simplified offsets and byte_counts."""
        if self.keyframe.is_contiguous:
            return self.data_offsets[:1], self.keyframe.is_contiguous[1:]
        return clean_offsets_counts(self.data_offsets, self.data_byte_counts)

    def is_contiguous(self):
        """Return offset and size of contiguous data, else None."""
        if self.keyframe.is_contiguous:
            return self.data_offsets[0], self.keyframe.is_contiguous[1]

    def is_memmappable(self, rgbonly, colormapped):
        """Return if page's image data in file can be memory-mapped."""
        return self.keyframe.is_memmappable(rgbonly, colormapped)

    def __getattr__(self, name):
        """Return attribute from keyframe."""
        if name in CONST.TIFF_FRAME_ATTRS:
            return getattr(self.keyframe, name)
        raise AttributeError("'%s' object has no attribute '%s'" %
                             (self.__class__.__name__, name))

    def __str__(self):
        """Return string containing information about frame."""
        s = '  '.join(s for s in (
            'x'.join(str(i) for i in self.shape),
        return "TiffFrame %i  %s" % (self.index, s)

    def info(self, verbose=True):
        """Return string with detailed information about frame."""
        result = [str(self)]
        names = ['data_offsets', 'data_byte_count'] + [
            CONST.TIFF_TAGS[i][0] for i in sorted(CONST.TIFF_FRAME_TAGS)]
        for name in names:
            if getattr(self, name, False):
                result.append('%s %s' % (name, getattr(self, name)))
        return '\n'.join(line[:CONST.PRINT_LINE_WIDTH] for line in result)

    def is_shaped(self):
        """Return description containing array shape if exists, else None."""
        d = getattr(self, 'image_description', False)
        if d:
            if b'"shape":' in d or d.startswith(b'shape=('):
                return d
            d = getattr(self, 'image_description_1', False)
            if d and (b'"shape":' in d or d.startswith(b'shape=(')):
                return d

class TiffTag(object):
    """TIFF tag structure.

    name : string
        Attribute name of tag.
    code : int
        Decimal code of tag.
    dtype : str
        Datatype of tag data. One of TIFF_DATA_TYPES.
    count : int
        Number of values.
    value : various types
        Tag data as Python object.
    value_offset : int
        Location of value in file.
    offset : int
        Location of tag in file.

    All attributes are read-only.

    __slots__ = ('code', 'name', 'count', 'dtype', 'offset',
                 'value', 'value_offset', '_value', '_type')

    class Error(Exception):

    def __init__(self, parent, **kwargs):
        """Initialize instance from file.

        File handle must be at offset to a TIFF tag header.

        fh = parent.filehandle
        byteorder = parent.byteorder
        unpack = struct.unpack
        tiff_tags = CONST.TIFF_TAGS
        tiff_custom_tags = CONST.TIFF_CUSTOM_TAGS

        self.offset = fh.tell()
        self.value_offset = self.offset + parent.offset_size + 4

        data = fh.read(parent.tag_size)
        code, dtype = unpack(parent.tag_format1, data[:4])
        count, value = unpack(parent.tag_format2, data[4:])
        self._value = value
        self._type = dtype

        if code in tiff_tags:
            name, _, _, count_, _ = tiff_tags[code]
            if count_ and count_ != count:
                count = count_
                warnings.warn("incorrect count for tag '%s'" % name)
        elif code in tiff_custom_tags:
            name = tiff_custom_tags[code][0]
            name = str(code)

            dtype = CONST.TIFF_DATA_TYPES[self._type]
        except KeyError:
            raise TiffTag.Error("unknown tag data type %i" % self._type)

        fmt = '%s%i%s' % (byteorder, count * int(dtype[0]), dtype[1])
        size = struct.calcsize(fmt)
        if size > parent.offset_size or code in tiff_custom_tags:
            pos = fh.tell()
            self.value_offset = offset = unpack(parent.offset_format, value)[0]
            if offset < 8 or offset > fh.size - size:
                raise TiffTag.Error("invalid tag value offset")
            # if offset % 2:
            #     warnings.warn("tag value does not begin on word boundary")
            if code in tiff_custom_tags:
                readfunc = tiff_custom_tags[code][1]
                value = readfunc(fh, byteorder, dtype, count)
            elif code in tiff_tags or dtype[-1] == 's':
                value = unpack(fmt, fh.read(size))
                value = read_numpy(fh, byteorder, dtype, count)
            value = unpack(fmt, value[:size])

        if code not in tiff_custom_tags and code not in CONST.TIFF_TAGS_TUPLE:
            if len(value) == 1:
                value = value[0]

        if (self._type != 7 and dtype[-1] == 's' and isinstance(value, bytes)):
            # TIFF ASCII fields can contain multiple strings,
            #   each terminated with a NUL
            value = stripascii(value)

        self.code = code
        self.name = name
        self.dtype = dtype
        self.count = count
        self.value = value

    def _fix_lsm_bitspersample(self, parent):
        """Correct LSM bitspersample tag.

        Old LSM writers may use a separate region for two 16-bit values,
        although they fit into the tag value element of the tag.

        if self.code == 258 and self.count == 2:
            # TODO: test this case; need example file
            warnings.warn("correcting LSM bitspersample tag")
            tof = parent.offset_format[parent.offset_size]
            self.value_offset = struct.unpack(tof, self._value)[0]
            self.value = struct.unpack("<HH", parent.filehandle.read(4))

    def as_str(self):
        """Return value as human readable string."""
        return ((str(self.value).split('\n', 1)[0]) if (self._type != 7)
                else '<undefined>')

    def __str__(self):
        """Return string containing information about tag."""
        value = ((str(self.value).split('\n', 1)[0]) if (self._type != 7)
                 else '<undefined>')
        tcode = "%i%s" % (self.count * int(self.dtype[0]), self.dtype[1])
        line = "TiffTag %i  %s  %s  %s" % (self.code, self.name, tcode, value)
        return line

class TiffPageSeries(object):
    """Series of TIFF pages with compatible shape and data type.

    pages : list of TiffPage
        Sequence of TiffPages in series.
    dtype : numpy.dtype or str
        Data type of the image array in series.
    shape : tuple
        Dimensions of the image array in series.
    axes : str
        Labels of axes in shape. See TiffPage.axes.
    offset : int or None
        Position of image data in file if memory-mappable, else None.

    def __init__(self, pages, shape, dtype, axes, parent=None):
        self.index = 0
        self.pages = pages
        self.shape = tuple(shape)
        self.axes = ''.join(axes)
        self.dtype = numpy.dtype(dtype)
        if parent:
            self.parent = parent
        elif len(pages):
            self.parent = pages[0].parent
            self.parent = None

    def asarray(self, memmap=False):
        """Return image data from series of TIFF pages as numpy array.

        memmap : bool
            If True, return an array stored in a binary file on disk
            if possible.

        if self.parent:
            return self.parent.asarray(series=self, memmap=memmap)

    def offset(self):
        """Return offset to memory-mappable data in page series."""
        if not self.pages:

        rgbonly = False
        pos = 0
        for page in self.pages:
            if page is None:
            if not page.is_memmappable(rgbonly, page.is_indexed):
            if not pos:
                pos = page.is_contiguous[0] + page.is_contiguous[1]
            if pos != page.is_contiguous[0]:
            pos += page.is_contiguous[1]

        offset = self.pages[0].is_contiguous[0]
        if (pos != offset + product(self.shape) * self.dtype.itemsize and
                not self.pages[0].is_imagej):
        return offset

    def __len__(self):
        """Return number of TiffPages in series."""
        return len(self.pages)

    def __getitem__(self, key):
        """Return specified TiffPage."""
        return self.pages[key]

    def __iter__(self):
        """Return iterator over TiffPages in series."""
        return iter(self.pages)

    def __str__(self):
        """Return string with information about series."""
        s = '  '.join(s for s in (
            'x'.join(str(i) for i in self.shape),
            '%i pages' % len(self.pages),
            ('memmap-offset=%i' % self.offset) if self.offset else
            'not mem-mappable'))
        return 'TiffPageSeries %i  %s' % (self.index, s)

class TiffSequence(object):
    """Sequence of image files.

    The data shape and dtype of all files must match.

    files : list
        List of file names.
    shape : tuple
        Shape of image sequence.
    axes : str
        Labels of axes in shape.

    >>> imsave('temp_C001T001.tif', numpy.random.rand(64, 64))
    >>> imsave('temp_C001T002.tif', numpy.random.rand(64, 64))
    >>> tifs = TiffSequence("temp_C001*.tif")
    >>> tifs.shape
    (1, 2)
    >>> tifs.axes
    >>> data = tifs.asarray()
    >>> data.shape
    (1, 2, 64, 64)

    _patterns = {
        'axes': r"""
            # matches Olympus OIF and Leica TIFF series

    class ParseError(Exception):

    def __init__(self, files, imread=TiffFile, pattern='axes',
                 *args, **kwargs):
        """Initialize instance from multiple files.

        files : str, or sequence of str
            Glob pattern or sequence of file names.
            Binary streams are not supported.
        imread : function or class
            Image read function or class with asarray function returning numpy
            array from single file.
        pattern : str
            Regular expression pattern that matches axes names and sequence
            indices in file names.
            By default this matches Olympus OIF and Leica TIFF series.

        if isinstance(files, basestring):
            files = natural_sorted(glob.glob(files))
        files = list(files)
        if not files:
            raise ValueError("no files found")
        if not isinstance(files[0], basestring):
            raise ValueError("not a file name")
        self.files = files

        if hasattr(imread, 'asarray'):
            # redefine imread
            _imread = imread

            def imread(fname, *args, **kwargs):
                with _imread(fname) as im:
                    return im.asarray(*args, **kwargs)

        self.imread = imread

        self.pattern = self._patterns.get(pattern, pattern)
            if not self.axes:
                self.axes = 'I'
        except self.ParseError:
            self.axes = 'I'
            self.shape = (len(files),)
            self._start_index = (0,)
            self._indices = tuple((i,) for i in range(len(files)))

    def __str__(self):
        """Return string with information about image sequence."""
        return "\n".join([
            '* files: %i' % len(self.files),
            '* axes: %s' % self.axes,
            '* shape: %s' % str(self.shape)])

    def __len__(self):
        return len(self.files)

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_value, traceback):

    def close(self):

    def asarray(self, memmap=False, tempdir=None, *args, **kwargs):
        """Read image data from all files and return as single numpy array.

        If memmap is True, return an array stored in a binary file on disk.
        The args and kwargs parameters are passed to the imread function.

        Raise IndexError or ValueError if image shapes do not match.

        im = self.imread(self.files[0], *args, **kwargs)
        shape = self.shape + im.shape
        if memmap:
            with tempfile.NamedTemporaryFile(dir=tempdir) as fh:
                result = numpy.memmap(fh, dtype=im.dtype, shape=shape)
            result = numpy.zeros(shape, dtype=im.dtype)
        result = result.reshape(-1, *im.shape)
        for index, fname in zip(self._indices, self.files):
            index = [i-j for i, j in zip(index, self._start_index)]
            index = numpy.ravel_multi_index(index, self.shape)
            im = self.imread(fname, *args, **kwargs)
            result[index] = im
        result.shape = shape
        return result

    def _parse(self):
        """Get axes and shape from file names."""
        if not self.pattern:
            raise self.ParseError("invalid pattern")
        pattern = re.compile(self.pattern, re.IGNORECASE | re.VERBOSE)
        matches = pattern.findall(self.files[0])
        if not matches:
            raise self.ParseError("pattern does not match file names")
        matches = matches[-1]
        if len(matches) % 2:
            raise self.ParseError("pattern does not match axis name and index")
        axes = ''.join(m for m in matches[::2] if m)
        if not axes:
            raise self.ParseError("pattern does not match file names")

        indices = []
        for fname in self.files:
            matches = pattern.findall(fname)[-1]
            if axes != ''.join(m for m in matches[::2] if m):
                raise ValueError("axes do not match within the image sequence")
            indices.append([int(m) for m in matches[1::2] if m])
        shape = tuple(numpy.max(indices, axis=0))
        start_index = tuple(numpy.min(indices, axis=0))
        shape = tuple(i-j+1 for i, j in zip(shape, start_index))
        if product(shape) != len(self.files):
            warnings.warn("files are missing. Missing data are zeroed")

        self.axes = axes.upper()
        self.shape = shape
        self._indices = indices
        self._start_index = start_index

class FileHandle(object):
    """Binary file handle.

    A limited, special purpose file handler that can:

    * handle embedded files (for CZI within CZI files)
    * re-open closed files (for multi file formats, such as OME-TIFF)
    * read and write numpy arrays and records from file like objects

    Only 'rb' and 'wb' modes are supported. Concurrently reading and writing
    of the same stream is untested.

    When initialized from another file handle, do not use it unless this
    FileHandle is closed.

    name : str
        Name of the file.
    path : str
        Absolute path to file.
    size : int
        Size of file in bytes.
    is_file : bool
        If True, file has a filno and can be memory-mapped.

    All attributes are read-only.

    __slots__ = ('_fh', '_file', '_mode', '_name', '_dir',
                 '_offset', '_size', '_close', 'is_file')

    def __init__(self, file, mode='rb', name=None, offset=None, size=None):
        """Initialize file handle from file name or another file handle.

        file : str, binary stream, or FileHandle
            File name or seekable binary stream, such as a open file
            or BytesIO.
        mode : str
            File open mode in case 'file' is a file name. Must be 'rb' or 'wb'.
        name : str
            Optional name of file in case 'file' is a binary stream.
        offset : int
            Optional start position of embedded file. By default this is
            the current file position.
        size : int
            Optional size of embedded file. By default this is the number
            of bytes from the 'offset' to the end of the file.

        self._fh = None
        self._file = file
        self._mode = mode
        self._name = name
        self._dir = ''
        self._offset = offset
        self._size = size
        self._close = True
        self.is_file = False

    def open(self):
        """Open or re-open file."""
        if self._fh:
            return  # file is open

        if isinstance(self._file, basestring):
            # file name
            self._file = os.path.realpath(self._file)
            self._dir, self._name = os.path.split(self._file)
            self._fh = open(self._file, self._mode)
            self._close = True
            if self._offset is None:
                self._offset = 0
        elif isinstance(self._file, FileHandle):
            # FileHandle
            self._fh = self._file._fh
            if self._offset is None:
                self._offset = 0
            self._offset += self._file._offset
            self._close = False
            if not self._name:
                if self._offset:
                    name, ext = os.path.splitext(self._file._name)
                    self._name = "%s@%i%s" % (name, self._offset, ext)
                    self._name = self._file._name
            if self._mode and self._mode != self._file._mode:
                raise ValueError('FileHandle has wrong mode')
            self._mode = self._file._mode
            self._dir = self._file._dir
        elif hasattr(self._file, 'seek'):
            # binary stream: open file, BytesIO
            except Exception:
                raise ValueError("binary stream is not seekable")
            self._fh = self._file
            if self._offset is None:
                self._offset = self._file.tell()
            self._close = False
            if not self._name:
                    self._dir, self._name = os.path.split(self._fh.name)
                except AttributeError:
                    self._name = "Unnamed binary stream"
                self._mode = self._fh.mode
            except AttributeError:
            raise ValueError("The first parameter must be a file name, "
                             "seekable binary stream, or FileHandle")

        if self._offset:

        if self._size is None:
            pos = self._fh.tell()
            self._fh.seek(self._offset, 2)
            self._size = self._fh.tell()

            self.is_file = True
        except Exception:
            self.is_file = False

    def read(self, size=-1):
        """Read 'size' bytes from file, or until EOF is reached."""
        if size < 0 and self._offset:
            size = self._size
        return self._fh.read(size)

    def write(self, bytestring):
        """Write bytestring to file."""
        return self._fh.write(bytestring)

    def flush(self):
        """Flush write buffers if applicable."""
        return self._fh.flush()

    def memmap_array(self, dtype, shape, offset=0, mode='r', order='C'):
        """Return numpy.memmap of data stored in file."""
        if not self.is_file:
            raise ValueError("Can not memory-map file without fileno")
        return numpy.memmap(self._fh, dtype=dtype, mode=mode,
                            offset=self._offset + offset,
                            shape=shape, order=order)

    def read_array(self, dtype, count=-1, sep=""):
        """Return numpy array from file.

        Work around numpy issue #2230, "numpy.fromfile does not accept
        StringIO object" https://github.com/numpy/numpy/issues/2230.

            return numpy.fromfile(self._fh, dtype, count, sep)
        except IOError:
            if count < 0:
                size = self._size
                size = count * numpy.dtype(dtype).itemsize
            data = self._fh.read(size)
            return numpy.fromstring(data, dtype, count, sep)

    def read_record(self, dtype, shape=1, byteorder=None):
        """Return numpy record from file."""
            rec = numpy.rec.fromfile(self._fh, dtype, shape,
        except Exception:
            dtype = numpy.dtype(dtype)
            if shape is None:
                shape = self._size // dtype.itemsize
            size = product(sequence(shape)) * dtype.itemsize
            data = self._fh.read(size)
            return numpy.rec.fromstring(data, dtype, shape,
        return rec[0] if shape == 1 else rec

    def write_array(self, data):
        """Write numpy array to binary file."""
        except Exception:
            # BytesIO

    def tell(self):
        """Return file's current position."""
        return self._fh.tell() - self._offset

    def seek(self, offset, whence=0):
        """Set file's current position."""
        if self._offset:
            if whence == 0:
                self._fh.seek(self._offset + offset, whence)
            elif whence == 2 and self._size > 0:
                self._fh.seek(self._offset + self._size + offset, 0)
        self._fh.seek(offset, whence)

    def close(self):
        """Close file."""
        if self._close and self._fh:
            self._fh = None

    def __enter__(self):
        return self

    def __exit__(self, exc_type, exc_value, traceback):

    def __getattr__(self, name):
        """Return attribute from underlying file object."""
        if self._offset:
                "FileHandle: '%s' not implemented for embedded files" % name)
        return getattr(self._fh, name)

    def name(self):
        return self._name

    def dirname(self):
        return self._dir

    def path(self):
        return os.path.join(self._dir, self._name)

    def size(self):
        return self._size

    def closed(self):
        return self._fh is None

class LazyConst(object):
    """Class whose attributes are computed on first access from its methods."""
    def __init__(self, cls):
        self._cls = cls

    def __getattr__(self, name):
        func = getattr(self._cls, name)
        if not callable(func):
            return func
            value = func()
        except TypeError:
            # Python 2 unbound method
            value = func.__func__()
        setattr(self, name, value)
        return value

class CONST(object):
    """Namespace for module constants."""

    def TIFF_TAGS():
        # Map TIFF tag code to
        #        (attribute name, default value, type, count, validator)
        return {
            254: ('new_subfile_type', 0, 4, 1, CONST.TIFF_SUBFILE_TYPES),
            255: ('subfile_type', None, 3, 1,
                  {0: 'undefined', 1: 'image', 2: 'reduced_image', 3: 'page'}),
            256: ('image_width', None, 4, 1, None),
            257: ('image_length', None, 4, 1, None),
            258: ('bits_per_sample', 1, 3, None, None),
            259: ('compression', 1, 3, 1, CONST.TIFF_COMPESSIONS),
            262: ('photometric', None, 3, 1, CONST.TIFF_PHOTOMETRICS),
            266: ('fill_order', 1, 3, 1, {1: 'msb2lsb', 2: 'lsb2msb'}),
            269: ('document_name', None, 2, None, None),
            270: ('image_description', None, 2, None, None),
            271: ('make', None, 2, None, None),
            272: ('model', None, 2, None, None),
            273: ('strip_offsets', None, 4, None, None),
            274: ('orientation', 1, 3, 1, CONST.TIFF_ORIENTATIONS),
            277: ('samples_per_pixel', 1, 3, 1, None),
            278: ('rows_per_strip', 2**32-1, 4, 1, None),
            279: ('strip_byte_counts', None, 4, None, None),
            280: ('min_sample_value', None, 3, None, None),
            281: ('max_sample_value', None, 3, None, None),  # 2**bitspersample
            282: ('x_resolution', None, 5, 1, None),
            283: ('y_resolution', None, 5, 1, None),
            284: ('planar_configuration', 1, 3, 1,
                  {1: 'contig', 2: 'separate'}),
            285: ('page_name', None, 2, None, None),
            286: ('x_position', None, 5, 1, None),
            287: ('y_position', None, 5, 1, None),
            292: ('group3_options', 0, 4, 1, CONST.TIFF_GROUP3_OPTIONS),
            296: ('resolution_unit', 2, 4, 1,
                  {1: None, 2: 'inch', 3: 'centimeter'}),
            297: ('page_number', None, 3, 2, None),
            305: ('software', None, 2, None, None),
            306: ('datetime', None, 2, None, None),
            315: ('artist', None, 2, None, None),
            316: ('host_computer', None, 2, None, None),
            317: ('predictor', 1, 3, 1,
                  {1: None, 2: 'horizontal', 3: 'float'}),
            318: ('white_point', None, 5, 2, None),
            319: ('primary_chromaticities', None, 5, 6, None),
            320: ('color_map', None, 3, None, None),
            322: ('tile_width', None, 4, 1, None),
            323: ('tile_length', None, 4, 1, None),
            324: ('tile_offsets', None, 4, None, None),
            325: ('tile_byte_counts', None, 4, None, None),
            330: ('sub_ifds', None, 4, None, None),
            338: ('extra_samples', None, 3, None,
                  {0: 'unspecified', 1: 'assocalpha', 2: 'unassalpha'}),
            339: ('sample_format', 1, 3, None, CONST.TIFF_SAMPLE_FORMATS),
            340: ('smin_sample_value', None, None, None, None),
            341: ('smax_sample_value', None, None, None, None),
            346: ('indexed', 0, 3, 1, None),
            347: ('jpeg_tables', None, 7, None, None),
            530: ('ycbcr_subsampling', (1, 1), 3, 2, None),
            531: ('ycbcr_positioning', (1, 1), 3, 1, None),
            532: ('reference_black_white', None, 5, 1, None),
            4864: ('andor', None, 2, None, None),
            32995: ('sgi_matteing', None, None, 1, None),  # use extra_samples
            32996: ('sgi_datatype', None, None, None, None),  # "
            32997: ('image_depth', 1, 4, 1, None),
            32998: ('tile_depth', None, 4, 1, None),
            33432: ('copyright', None, 1, None, None),
            33445: ('md_file_tag', None, 4, 1, None),
            33446: ('md_scale_pixel', None, 5, 1, None),
            33447: ('md_color_table', None, 3, None, None),
            33448: ('md_lab_name', None, 2, None, None),
            33449: ('md_sample_info', None, 2, None, None),
            33450: ('md_prep_date', None, 2, None, None),
            33451: ('md_prep_time', None, 2, None, None),
            33452: ('md_file_units', None, 2, None, None),
            33550: ('model_pixel_scale', None, 12, 3, None),
            33922: ('model_tie_point', None, 12, None, None),
            34665: ('exif_ifd', None, None, 1, None),
            34735: ('geo_key_directory', None, 3, None, None),
            34736: ('geo_double_params', None, 12, None, None),
            34737: ('geo_ascii_params', None, 2, None, None),
            34853: ('gps_ifd', None, None, 1, None),
            37510: ('user_comment', None, None, None, None),
            42112: ('gdal_metadata', None, 2, None, None),
            42113: ('gdal_nodata', None, 2, None, None),
            50289: ('mc_xy_position', None, 12, 2, None),
            50290: ('mc_z_position', None, 12, 1, None),
            50291: ('mc_xy_calibration', None, 12, 3, None),
            50292: ('mc_lens_lem_na_n', None, 12, 3, None),
            50293: ('mc_channel_name', None, 1, None, None),
            50294: ('mc_ex_wavelength', None, 12, 1, None),
            50295: ('mc_time_stamp', None, 12, 1, None),
            50838: ('imagej_byte_counts', None, None, None, None),
            51023: ('fibics_xml', None, 2, None, None),
            65200: ('flex_xml', None, 2, None, None),

        # Map custom TIFF tag codes to attribute names and import functions
        return {
            700: ('xmp', read_bytes),
            34377: ('photoshop', read_numpy),
            33723: ('iptc', read_bytes),
            34675: ('icc_profile', read_bytes),
            33628: ('uic1tag', read_uic1tag),  # Universal Imaging Corp STK
            33629: ('uic2tag', read_uic2tag),
            33630: ('uic3tag', read_uic3tag),
            33631: ('uic4tag', read_uic4tag),
            34118: ('sem_metadata', read_sem_metadata),  # Zeiss SEM
            34361: ('mm_header', read_mm_header),  # Olympus FluoView
            34362: ('mm_stamp', read_mm_stamp),
            34386: ('mm_user_block', read_bytes),
            34412: ('cz_lsm_info', read_cz_lsm_info),  # Carl Zeiss LSM
            34680: ('sfeg_metadata', read_fei_metadata),  # S-FEG
            34682: ('helios_metadata', read_fei_metadata),  # Helios NanoLab
            37706: ('tvips_metadata', read_tvips_header),  # TVIPS EMMENU
            43314: ('nih_image_header', read_nih_image_header),
            # 40001: ('mc_ipwinscal', read_bytes),
            40100: ('mc_id_old', read_bytes),
            50288: ('mc_id', read_bytes),
            50296: ('mc_frame_properties', read_bytes),
            50839: ('imagej_metadata', read_bytes),
            51123: ('micromanager_metadata', read_json),

    def TIFF_TAGS_NAMES():
        return dict((v[0], c) for c, v in CONST.TIFF_TAGS.items())

        # Tag default values
        return dict((name, validate[default] if validate else default)
                    for code, (name, default, dtype, count, validate)
                    in CONST.TIFF_TAGS.items()
                    if default is not None)

    def TIFF_TAGS_TUPLE():
        # Tags whose values must be stored as tuples
        return set((273, 279, 324, 325, 530, 531))

    def TIFF_DATA_TYPES():
        # Map TIFF data types to numpy
        return {
            1: '1B',   # BYTE 8-bit unsigned integer.
            2: '1s',   # ASCII 8-bit byte that contains a 7-bit ASCII code;
                       #   the last byte must be NULL (binary zero).
            3: '1H',   # SHORT 16-bit (2-byte) unsigned integer
            4: '1I',   # LONG 32-bit (4-byte) unsigned integer.
            5: '2I',   # RATIONAL Two LONGs: the first represents the numerator
                       #   of a fraction; the second, the denominator.
            6: '1b',   # SBYTE An 8-bit signed (twos-complement) integer.
            7: '1s',   # UNDEFINED An 8-bit byte that may contain anything,
                       #   depending on the definition of the field.
            8: '1h',   # SSHORT A 16-bit (2-byte) signed (twos-complement)
                       #   integer.
            9: '1i',   # SLONG A 32-bit (4-byte) signed (twos-complement)
                       #   integer.
            10: '2i',  # SRATIONAL Two SLONGs: the first represents the
                       #   numerator of a fraction, the second the denominator.
            11: '1f',  # FLOAT Single precision (4-byte) IEEE format.
            12: '1d',  # DOUBLE Double precision (8-byte) IEEE format.
            13: '1I',  # IFD unsigned 4 byte IFD offset.
            # 14: '',   # UNICODE
            # 15: '',   # COMPLEX
            16: '1Q',  # LONG8 unsigned 8 byte integer (BigTiff)
            17: '1q',  # SLONG8 signed 8 byte integer (BigTiff)
            18: '1Q',  # IFD8 unsigned 8 byte IFD offset (BigTiff)

        # Map numpy data types to TIFF
        return {'B': 1, 's': 2, 'H': 3, 'I': 4, '2I': 5, 'b': 6,
                'h': 8, 'i': 9, '2i': 10, 'f': 11, 'd': 12, 'Q': 16, 'q': 17}

        return {
            1: 'uint',
            2: 'int',
            3: 'float',
            # 4: 'void',
            # 5: 'complex_int',
            6: 'complex',

        # Map TIFF sample formats and bits per sample to numpy dtype
        return {
            ('uint', 1): '?',  # bitmap
            ('uint', 2): 'B',
            ('uint', 3): 'B',
            ('uint', 4): 'B',
            ('uint', 5): 'B',
            ('uint', 6): 'B',
            ('uint', 7): 'B',
            ('uint', 8): 'B',
            ('uint', 9): 'H',
            ('uint', 10): 'H',
            ('uint', 11): 'H',
            ('uint', 12): 'H',
            ('uint', 13): 'H',
            ('uint', 14): 'H',
            ('uint', 15): 'H',
            ('uint', 16): 'H',
            ('uint', 17): 'I',
            ('uint', 18): 'I',
            ('uint', 19): 'I',
            ('uint', 20): 'I',
            ('uint', 21): 'I',
            ('uint', 22): 'I',
            ('uint', 23): 'I',
            ('uint', 24): 'I',
            ('uint', 25): 'I',
            ('uint', 26): 'I',
            ('uint', 27): 'I',
            ('uint', 28): 'I',
            ('uint', 29): 'I',
            ('uint', 30): 'I',
            ('uint', 31): 'I',
            ('uint', 32): 'I',
            ('uint', 64): 'Q',
            ('int', 8): 'b',
            ('int', 16): 'h',
            ('int', 32): 'i',
            ('int', 64): 'q',
            ('float', 16): 'e',
            ('float', 32): 'f',
            ('float', 64): 'd',
            ('complex', 64): 'F',
            ('complex', 128): 'D',
            ('uint', (5, 6, 5)): 'B',

        return {
            0: 'miniswhite',
            1: 'minisblack',
            2: 'rgb',
            3: 'palette',
            4: 'mask',
            5: 'separated',  # CMYK
            6: 'ycbcr',
            8: 'cielab',
            9: 'icclab',
            10: 'itulab',
            32803: 'cfa',  # Color Filter Array
            32844: 'logl',
            32845: 'logluv',
            34892: 'linear_raw'

        return {
            1: None,
            2: 'ccittrle',
            3: 'ccittfax3',
            4: 'ccittfax4',
            5: 'lzw',
            6: 'ojpeg',
            7: 'jpeg',
            8: 'adobe_deflate',
            9: 't85',
            10: 't43',
            32766: 'next',
            32771: 'ccittrlew',
            32773: 'packbits',
            32809: 'thunderscan',
            32895: 'it8ctpad',
            32896: 'it8lw',
            32897: 'it8mp',
            32898: 'it8bl',
            32908: 'pixarfilm',
            32909: 'pixarlog',
            32946: 'deflate',
            32947: 'dcs',
            33003: 'aperio_jp2000_ycbcr',
            33005: 'aperio_jp2000_rgb',
            34661: 'jbig',
            34676: 'sgilog',
            34677: 'sgilog24',
            34712: 'jp2000',
            34713: 'nef',
            34925: 'lzma',

        decompressors = {
            None: lambda x: x,
            'adobe_deflate': zlib.decompress,
            'deflate': zlib.decompress,
            'packbits': decode_packbits,
            'lzw': decode_lzw,
            # 'jpeg': decode_jpeg
        if lzma:
            decompressors['lzma'] = lzma.decompress
        return decompressors

        return {
            1: 'top_left',
            2: 'top_right',
            3: 'bottom_right',
            4: 'bottom_left',
            5: 'left_top',
            6: 'right_top',
            7: 'right_bottom',
            8: 'left_bottom',

    class TIFF_SUBFILE_TYPES(object):
        def __getitem__(self, key):
            result = []
            if key & 1:
            if key & 2:
            if key & 4:
            return tuple(result)

    class TIFF_GROUP3_OPTIONS(object):
        def __getitem__(self, key):
            result = []
            if key & 1:
            if key & 2:
            if key & 4:
            return tuple(result)

    def TIFF_FRAME_TAGS():
        # Tags to be read from file by TiffFrame
        return set((273, 279, 324, 325, 297, 285, 306, 270))

        # Attributes that a TiffFrame shares with keyframe
        return set('shape ndim size dtype axes is_indexed'.split())
        #          'compression photometric is_reduced is_rgb is_tiled '
        #          'image_width image_length image_depth '
        #          'samples_per_pixel bits_per_sample sample_format '

    def TIFF_PAGE_ATTRS():
        # TiffPage attributes
        return set('reduced sgi tiled indexed contiguous '

    def TIFF_FILE_ATTRS():
        # TiffFile 'is_\*' attributes derived from first TiffPage
        return set('shaped ome imagej stk lsm fluoview nih vista '
                   'micromanager metaseries mdgel mediacy tvips fei '
                   'sem scn svs scanimage andor'.split())

        # TIFF file extensions
        return tuple('tif tiff ome.tif lsm stk '
                     'gel seq svs bif tf8 tf2 btf'.split())

        # String for use in Windows File Open box
        return [("%s files" % ext.upper(), "*.%s" % ext)
                for ext in CONST.TIFF_FILE_EXTENSIONS] + [("allfiles", "*")]

    def AXES_LABELS():
        # TODO: is there a standard for character axes labels?
        axes = {
            'X': 'width',
            'Y': 'height',
            'Z': 'depth',
            'S': 'sample',  # rgb(a)
            'I': 'series',  # general sequence, plane, page, IFD
            'T': 'time',
            'C': 'channel',  # color, emission wavelength
            'A': 'angle',
            'P': 'phase',  # formerly F    # P is Position in LSM!
            'R': 'tile',  # region, point, mosaic
            'H': 'lifetime',  # histogram
            'E': 'lambda',  # excitation wavelength
            'L': 'exposure',  # lux
            'V': 'event',
            'Q': 'other',
            'M': 'mosaic',  # LSM 6
        axes.update(dict((v, k) for k, v in axes.items()))
        return axes

    def ANDOR_TAGS():
        # Andor Technology tags #4864 - 5030
        # TODO: obtain specification for Andor tags
        return {
            4864: 'id',
            4869: 'temperature',
            4876: 'exposure_time',
            4878: 'kinetic_cycle_time',
            4879: 'accumulations',
            4881: 'acquisition_cycle_time',
            4882: 'readout_time',
            4884: 'photon_counting',
            4885: 'em_dac_level',
            4890: 'frames',
            4896: 'horizontal_flip',
            4897: 'vertical_flip',
            4898: 'clockwise',
            4899: 'counter_clockwise',
            4904: 'vertical_clock_voltage',
            4905: 'vertical_shift_speed',
            4907: 'pre_amp_setting',
            4908: 'camera_serial',
            4911: 'actual_temperature',
            4912: 'baseline_clamp',
            4913: 'prescans',
            4914: 'model',
            4915: 'chip_size_x',
            4916: 'chip_size_y',
            4944: 'baseline_offset',
            4966: 'software_version',

        # Import functions for LSM_INFO sub-records
        return {
            'scan_info': read_cz_lsm_scan_info,
            'time_stamps': read_cz_lsm_time_stamps,
            'event_list': read_cz_lsm_event_list,
            'channel_colors': read_cz_lsm_floatpairs,
            'positions': read_cz_lsm_floatpairs,
            'tile_positions': read_cz_lsm_floatpairs,

    def CZ_LSM_INFO():
        return [
            ('magic_number', 'u4'),
            ('structure_size', 'i4'),
            ('dimension_x', 'i4'),
            ('dimension_y', 'i4'),
            ('dimension_z', 'i4'),
            ('dimension_channels', 'i4'),
            ('dimension_time', 'i4'),
            ('data_type', 'i4'),  # CZ_DATA_TYPES
            ('thumbnail_x', 'i4'),
            ('thumbnail_y', 'i4'),
            ('voxel_size_x', 'f8'),
            ('voxel_size_y', 'f8'),
            ('voxel_size_z', 'f8'),
            ('origin_x', 'f8'),
            ('origin_y', 'f8'),
            ('origin_z', 'f8'),
            ('scan_type', 'u2'),
            ('spectral_scan', 'u2'),
            ('type_of_data', 'u4'),  # CZ_TYPE_OF_DATA
            ('offset_vector_overlay', 'u4'),
            ('offset_input_lut', 'u4'),
            ('offset_output_lut', 'u4'),
            ('offset_channel_colors', 'u4'),
            ('time_interval', 'f8'),
            ('offset_channel_data_types', 'u4'),
            ('offset_scan_info', 'u4'),  # CZ_LSM_SCAN_INFO
            ('offset_ks_data', 'u4'),
            ('offset_time_stamps', 'u4'),
            ('offset_event_list', 'u4'),
            ('offset_roi', 'u4'),
            ('offset_bleach_roi', 'u4'),
            ('offset_next_recording', 'u4'),
            # LSM 2.0 ends here
            ('display_aspect_x', 'f8'),
            ('display_aspect_y', 'f8'),
            ('display_aspect_z', 'f8'),
            ('display_aspect_time', 'f8'),
            ('offset_mean_of_roi_overlay', 'u4'),
            ('offset_topo_isoline_overlay', 'u4'),
            ('offset_topo_profile_overlay', 'u4'),
            ('offset_linescan_overlay', 'u4'),
            ('offset_toolbar_flags', 'u4'),
            ('offset_channel_wavelength', 'u4'),
            ('offset_channel_factors', 'u4'),
            ('objective_sphere_correction', 'f8'),
            ('offset_unmix_parameters', 'u4'),
            # LSM 3.2, 4.0 end here
            ('offset_acquisition_parameters', 'u4'),
            ('offset_characteristics', 'u4'),
            ('offset_palette', 'u4'),
            ('time_difference_x', 'f8'),
            ('time_difference_y', 'f8'),
            ('time_difference_z', 'f8'),
            ('internal_use_1', 'u4'),
            ('dimension_p', 'i4'),
            ('dimension_m', 'i4'),
            ('dimensions_reserved', '16i4'),
            ('offset_tile_positions', 'u4'),
            ('reserved_1', '9u4'),
            ('offset_positions', 'u4'),
            ('reserved_2', '21u4'),  # must be 0

    def CZ_SCAN_TYPES():
        # Map cz_lsm_info.scan_type to dimension order
        return {
            0: 'XYZCT',  # x-y-z scan
            1: 'XYZCT',  # z scan (x-z plane)
            2: 'XYZCT',  # line scan
            3: 'XYTCZ',  # time series x-y
            4: 'XYZTC',  # time series x-z
            5: 'XYTCZ',  # time series 'Mean of ROIs'
            6: 'XYZTC',  # time series x-y-z
            7: 'XYCTZ',  # spline scan
            8: 'XYCZT',  # spline scan x-z
            9: 'XYTCZ',  # time series spline plane x-z
            10: 'XYZCT',  # point mode

    def CZ_DIMENSIONS():
        # Map dimension codes to cz_lsm_info attribute
        return {
            'X': 'dimension_x',
            'Y': 'dimension_y',
            'Z': 'dimension_z',
            'C': 'dimension_channels',
            'T': 'dimension_time',
            'P': 'dimension_p',
            'M': 'dimension_m',

    def CZ_DATA_TYPES():
        # Description of cz_lsm_info.data_type
        return {
            0: 'varying data types',
            1: '8 bit unsigned integer',
            2: '12 bit unsigned integer',
            5: '32 bit float',

    def CZ_TYPE_OF_DATA():
        # Description of cz_lsm_info.type_of_data
        return {
            0: 'Original scan data',
            1: 'Calculated data',
            2: '3D reconstruction',
            3: 'Topography height map',

        return {
            0x20000000: "tracks",
            0x30000000: "lasers",
            0x60000000: "detection_channels",
            0x80000000: "illumination_channels",
            0xa0000000: "beam_splitters",
            0xc0000000: "data_channels",
            0x11000000: "timers",
            0x13000000: "markers",

        return {
            # 0x10000000: "recording",
            0x40000000: "track",
            0x50000000: "laser",
            0x70000000: "detection_channel",
            0x90000000: "illumination_channel",
            0xb0000000: "beam_splitter",
            0xd0000000: "data_channel",
            0x12000000: "timer",
            0x14000000: "marker",

        return {
            # recording
            0x10000001: "name",
            0x10000002: "description",
            0x10000003: "notes",
            0x10000004: "objective",
            0x10000005: "processing_summary",
            0x10000006: "special_scan_mode",
            0x10000007: "scan_type",
            0x10000008: "scan_mode",
            0x10000009: "number_of_stacks",
            0x1000000a: "lines_per_plane",
            0x1000000b: "samples_per_line",
            0x1000000c: "planes_per_volume",
            0x1000000d: "images_width",
            0x1000000e: "images_height",
            0x1000000f: "images_number_planes",
            0x10000010: "images_number_stacks",
            0x10000011: "images_number_channels",
            0x10000012: "linscan_xy_size",
            0x10000013: "scan_direction",
            0x10000014: "time_series",
            0x10000015: "original_scan_data",
            0x10000016: "zoom_x",
            0x10000017: "zoom_y",
            0x10000018: "zoom_z",
            0x10000019: "sample_0x",
            0x1000001a: "sample_0y",
            0x1000001b: "sample_0z",
            0x1000001c: "sample_spacing",
            0x1000001d: "line_spacing",
            0x1000001e: "plane_spacing",
            0x1000001f: "plane_width",
            0x10000020: "plane_height",
            0x10000021: "volume_depth",
            0x10000023: "nutation",
            0x10000034: "rotation",
            0x10000035: "precession",
            0x10000036: "sample_0time",
            0x10000037: "start_scan_trigger_in",
            0x10000038: "start_scan_trigger_out",
            0x10000039: "start_scan_event",
            0x10000040: "start_scan_time",
            0x10000041: "stop_scan_trigger_in",
            0x10000042: "stop_scan_trigger_out",
            0x10000043: "stop_scan_event",
            0x10000044: "stop_scan_time",
            0x10000045: "use_rois",
            0x10000046: "use_reduced_memory_rois",
            0x10000047: "user",
            0x10000048: "use_bc_correction",
            0x10000049: "position_bc_correction1",
            0x10000050: "position_bc_correction2",
            0x10000051: "interpolation_y",
            0x10000052: "camera_binning",
            0x10000053: "camera_supersampling",
            0x10000054: "camera_frame_width",
            0x10000055: "camera_frame_height",
            0x10000056: "camera_offset_x",
            0x10000057: "camera_offset_y",
            0x10000059: "rt_binning",
            0x1000005a: "rt_frame_width",
            0x1000005b: "rt_frame_height",
            0x1000005c: "rt_region_width",
            0x1000005d: "rt_region_height",
            0x1000005e: "rt_offset_x",
            0x1000005f: "rt_offset_y",
            0x10000060: "rt_zoom",
            0x10000061: "rt_line_period",
            0x10000062: "prescan",
            0x10000063: "scan_direction_z",
            # track
            0x40000001: "multiplex_type",  # 0 after line; 1 after frame
            0x40000002: "multiplex_order",
            0x40000003: "sampling_mode",  # 0 sample; 1 line avg; 2 frame avg
            0x40000004: "sampling_method",  # 1 mean; 2 sum
            0x40000005: "sampling_number",
            0x40000006: "acquire",
            0x40000007: "sample_observation_time",
            0x4000000b: "time_between_stacks",
            0x4000000c: "name",
            0x4000000d: "collimator1_name",
            0x4000000e: "collimator1_position",
            0x4000000f: "collimator2_name",
            0x40000010: "collimator2_position",
            0x40000011: "is_bleach_track",
            0x40000012: "is_bleach_after_scan_number",
            0x40000013: "bleach_scan_number",
            0x40000014: "trigger_in",
            0x40000015: "trigger_out",
            0x40000016: "is_ratio_track",
            0x40000017: "bleach_count",
            0x40000018: "spi_center_wavelength",
            0x40000019: "pixel_time",
            0x40000021: "condensor_frontlens",
            0x40000023: "field_stop_value",
            0x40000024: "id_condensor_aperture",
            0x40000025: "condensor_aperture",
            0x40000026: "id_condensor_revolver",
            0x40000027: "condensor_filter",
            0x40000028: "id_transmission_filter1",
            0x40000029: "id_transmission1",
            0x40000030: "id_transmission_filter2",
            0x40000031: "id_transmission2",
            0x40000032: "repeat_bleach",
            0x40000033: "enable_spot_bleach_pos",
            0x40000034: "spot_bleach_posx",
            0x40000035: "spot_bleach_posy",
            0x40000036: "spot_bleach_posz",
            0x40000037: "id_tubelens",
            0x40000038: "id_tubelens_position",
            0x40000039: "transmitted_light",
            0x4000003a: "reflected_light",
            0x4000003b: "simultan_grab_and_bleach",
            0x4000003c: "bleach_pixel_time",
            # laser
            0x50000001: "name",
            0x50000002: "acquire",
            0x50000003: "power",
            # detection_channel
            0x70000001: "integration_mode",
            0x70000002: "special_mode",
            0x70000003: "detector_gain_first",
            0x70000004: "detector_gain_last",
            0x70000005: "amplifier_gain_first",
            0x70000006: "amplifier_gain_last",
            0x70000007: "amplifier_offs_first",
            0x70000008: "amplifier_offs_last",
            0x70000009: "pinhole_diameter",
            0x7000000a: "counting_trigger",
            0x7000000b: "acquire",
            0x7000000c: "point_detector_name",
            0x7000000d: "amplifier_name",
            0x7000000e: "pinhole_name",
            0x7000000f: "filter_set_name",
            0x70000010: "filter_name",
            0x70000013: "integrator_name",
            0x70000014: "channel_name",
            0x70000015: "detector_gain_bc1",
            0x70000016: "detector_gain_bc2",
            0x70000017: "amplifier_gain_bc1",
            0x70000018: "amplifier_gain_bc2",
            0x70000019: "amplifier_offset_bc1",
            0x70000020: "amplifier_offset_bc2",
            0x70000021: "spectral_scan_channels",
            0x70000022: "spi_wavelength_start",
            0x70000023: "spi_wavelength_stop",
            0x70000026: "dye_name",
            0x70000027: "dye_folder",
            # illumination_channel
            0x90000001: "name",
            0x90000002: "power",
            0x90000003: "wavelength",
            0x90000004: "aquire",
            0x90000005: "detchannel_name",
            0x90000006: "power_bc1",
            0x90000007: "power_bc2",
            # beam_splitter
            0xb0000001: "filter_set",
            0xb0000002: "filter",
            0xb0000003: "name",
            # data_channel
            0xd0000001: "name",
            0xd0000003: "acquire",
            0xd0000004: "color",
            0xd0000005: "sample_type",
            0xd0000006: "bits_per_sample",
            0xd0000007: "ratio_type",
            0xd0000008: "ratio_track1",
            0xd0000009: "ratio_track2",
            0xd000000a: "ratio_channel1",
            0xd000000b: "ratio_channel2",
            0xd000000c: "ratio_const1",
            0xd000000d: "ratio_const2",
            0xd000000e: "ratio_const3",
            0xd000000f: "ratio_const4",
            0xd0000010: "ratio_const5",
            0xd0000011: "ratio_const6",
            0xd0000012: "ratio_first_images1",
            0xd0000013: "ratio_first_images2",
            0xd0000014: "dye_name",
            0xd0000015: "dye_folder",
            0xd0000016: "spectrum",
            0xd0000017: "acquire",
            # timer
            0x12000001: "name",
            0x12000002: "description",
            0x12000003: "interval",
            0x12000004: "trigger_in",
            0x12000005: "trigger_out",
            0x12000006: "activation_time",
            0x12000007: "activation_number",
            # marker
            0x14000001: "name",
            0x14000002: "description",
            0x14000003: "trigger_in",
            0x14000004: "trigger_out",

        return [
            ('fileid', 'a8'),
            ('nlines', 'i2'),
            ('pixelsperline', 'i2'),
            ('version', 'i2'),
            ('oldlutmode', 'i2'),
            ('oldncolors', 'i2'),
            ('colors', 'u1', (3, 32)),
            ('oldcolorstart', 'i2'),
            ('colorwidth', 'i2'),
            ('extracolors', 'u2', (6, 3)),
            ('nextracolors', 'i2'),
            ('foregroundindex', 'i2'),
            ('backgroundindex', 'i2'),
            ('xscale', 'f8'),
            ('_x0', 'i2'),
            ('_x1', 'i2'),
            ('units_t', 'i2'),  # NIH_UNITS_TYPE
            ('p1', [('x', 'i2'), ('y', 'i2')]),
            ('p2', [('x', 'i2'), ('y', 'i2')]),
            ('curvefit_t', 'i2'),  # NIH_CURVEFIT_TYPE
            ('ncoefficients', 'i2'),
            ('coeff', 'f8', 6),
            ('_um_len', 'u1'),
            ('um', 'a15'),
            ('_x2', 'u1'),
            ('binarypic', 'b1'),
            ('slicestart', 'i2'),
            ('sliceend', 'i2'),
            ('scalemagnification', 'f4'),
            ('nslices', 'i2'),
            ('slicespacing', 'f4'),
            ('currentslice', 'i2'),
            ('frameinterval', 'f4'),
            ('pixelaspectratio', 'f4'),
            ('colorstart', 'i2'),
            ('colorend', 'i2'),
            ('ncolors', 'i2'),
            ('fill1', '3u2'),
            ('fill2', '3u2'),
            ('colortable_t', 'u1'),  # NIH_COLORTABLE_TYPE
            ('lutmode_t', 'u1'),  # NIH_LUTMODE_TYPE
            ('invertedtable', 'b1'),
            ('zeroclip', 'b1'),
            ('_xunit_len', 'u1'),
            ('xunit', 'a11'),
            ('stacktype_t', 'i2'),  # NIH_STACKTYPE_TYPE

        return ('CustomTable', 'AppleDefault', 'Pseudo20', 'Pseudo32',
                'Rainbow', 'Fire1', 'Fire2', 'Ice', 'Grays', 'Spectrum')

        return ('PseudoColor', 'OldAppleDefault', 'OldSpectrum', 'GrayScale',
                'ColorLut', 'CustomGrayscale')

        return ('StraightLine', 'Poly2', 'Poly3', 'Poly4', 'Poly5', 'ExpoFit',
                'PowerFit', 'LogFit', 'RodbardFit', 'SpareFit1',
                'Uncalibrated', 'UncalibratedOD')

    def NIH_UNITS_TYPE():
        return ('Nanometers', 'Micrometers', 'Millimeters', 'Centimeters',
                'Meters', 'Kilometers', 'Inches', 'Feet', 'Miles', 'Pixels',

        return ('VolumeStack', 'RGBStack', 'MovieStack', 'HSVStack')

    def TVIPS_HEADER_V1():
        # TVIPS metadata from EMMENU Help file
        return [
            ('version', 'i4'),
            ('comment_v1', 'a80'),
            ('high_tension', 'i4'),
            ('spherical_aberration', 'i4'),
            ('illumination_aperture', 'i4'),
            ('magnification', 'i4'),
            ('post-magnification', 'i4'),
            ('focal_length', 'i4'),
            ('defocus', 'i4'),
            ('astigmatism', 'i4'),
            ('astigmatism_direction', 'i4'),
            ('biprism_voltage', 'i4'),
            ('specimen_tilt_angle', 'i4'),
            ('specimen_tilt_direction', 'i4'),
            ('illumination_tilt_direction', 'i4'),
            ('illumination_tilt_angle', 'i4'),
            ('image_mode', 'i4'),
            ('energy_spread', 'i4'),
            ('chromatic_aberration', 'i4'),
            ('shutter_type', 'i4'),
            ('defocus_spread', 'i4'),
            ('ccd_number', 'i4'),
            ('ccd_size', 'i4'),
            ('offset_x_v1', 'i4'),
            ('offset_y_v1', 'i4'),
            ('physical_pixel_size', 'i4'),
            ('binning', 'i4'),
            ('readout_speed', 'i4'),
            ('gain_v1', 'i4'),
            ('sensitivity_v1', 'i4'),
            ('exposure_time_v1', 'i4'),
            ('flat_corrected', 'i4'),
            ('dead_px_corrected', 'i4'),
            ('image_mean', 'i4'),
            ('image_std', 'i4'),
            ('displacement_x', 'i4'),
            ('displacement_y', 'i4'),
            ('date_v1', 'i4'),
            ('time_v1', 'i4'),
            ('image_min', 'i4'),
            ('image_max', 'i4'),
            ('image_statistics_quality', 'i4'),

    def TVIPS_HEADER_V2():
        return [
            ('image_name', 'V160'),  # utf16
            ('image_folder', 'V160'),
            ('image_size_x', 'i4'),
            ('image_size_y', 'i4'),
            ('image_size_z', 'i4'),
            ('image_size_e', 'i4'),
            ('image_data_type', 'i4'),
            ('date', 'i4'),
            ('time', 'i4'),
            ('comment', 'V1024'),
            ('image_history', 'V1024'),
            ('scaling', '16f4'),
            ('image_statistics', '16c16'),
            ('image_type', 'i4'),
            ('image_display_type', 'i4'),
            ('pixel_size_x', 'f4'),  # distance between two px in x, [nm]
            ('pixel_size_y', 'f4'),  # distance between two px in y, [nm]
            ('image_distance_z', 'f4'),
            ('image_distance_e', 'f4'),
            ('image_misc', '32f4'),
            ('tem_type', 'V160'),
            ('tem_high_tension', 'f4'),
            ('tem_aberrations', '32f4'),
            ('tem_energy', '32f4'),
            ('tem_mode', 'i4'),
            ('tem_magnification', 'f4'),
            ('tem_magnification_correction', 'f4'),
            ('post_magnification', 'f4'),
            ('tem_stage_type', 'i4'),
            ('tem_stage_position', '5f4'),  # x, y, z, a, b
            ('tem_image_shift', '2f4'),
            ('tem_beam_shift', '2f4'),
            ('tem_beam_tilt', '2f4'),
            ('tiling_parameters', '7f4'),  # 0: tiling? 1:x 2:y 3: max x
                                           # 4: max y 5: overlap x 6: overlap y
            ('tem_illumination', '3f4'),  # 0: spotsize 1: intensity
            ('tem_shutter', 'i4'),
            ('tem_misc', '32f4'),
            ('camera_type', 'V160'),
            ('physical_pixel_size_x', 'f4'),
            ('physical_pixel_size_y', 'f4'),
            ('offset_x', 'i4'),
            ('offset_y', 'i4'),
            ('binning_x', 'i4'),
            ('binning_y', 'i4'),
            ('exposure_time', 'f4'),
            ('gain', 'f4'),
            ('readout_rate', 'f4'),
            ('flatfield_description', 'V160'),
            ('sensitivity', 'f4'),
            ('dose', 'f4'),
            ('cam_misc', '32f4'),
            ('fei_microscope_information', 'V1024'),
            ('fei_specimen_information', 'V1024'),
            ('magic', 'u4'),

        # Olympus FluoView MM header
        MM_DIMENSION = [
            ('name', 'a16'),
            ('size', 'i4'),
            ('origin', 'f8'),
            ('resolution', 'f8'),
            ('unit', 'a64')]
        return [
            ('header_flag', 'i2'),
            ('image_type', 'u1'),
            ('image_name', 'a257'),
            ('offset_data', 'u4'),
            ('palette_size', 'i4'),
            ('offset_palette0', 'u4'),
            ('offset_palette1', 'u4'),
            ('comment_size', 'i4'),
            ('offset_comment', 'u4'),
            ('dimensions', MM_DIMENSION, 10),
            ('offset_position', 'u4'),
            ('map_type', 'i2'),
            ('map_min', 'f8'),
            ('map_max', 'f8'),
            ('min_value', 'f8'),
            ('max_value', 'f8'),
            ('offset_map', 'u4'),
            ('gamma', 'f8'),
            ('offset', 'f8'),
            ('gray_channel', MM_DIMENSION),
            ('offset_thumbnail', 'u4'),
            ('voice_field', 'i4'),
            ('offset_voice_field', 'u4'),

        # Map fluoview_mm_header.dimensions to axes characters
        return {
            b'X': 'X',
            b'Y': 'Y',
            b'Z': 'Z',
            b'T': 'T',
            b'CH': 'C',
            b'WAVELENGTH': 'C',
            b'TIME': 'T',
            b'XY': 'R',
            b'EVENT': 'V',
            b'EXPOSURE': 'L',

    def UIC_TAGS():
        # Map Universal Imaging Corporation MetaMorph internal tag ids to
        # name and type
        from fractions import Fraction

        return [
            ('auto_scale', int),
            ('min_scale', int),
            ('max_scale', int),
            ('spatial_calibration', int),
            ('x_calibration', Fraction),
            ('y_calibration', Fraction),
            ('calibration_units', str),
            ('name', str),
            ('thresh_state', int),
            ('thresh_state_red', int),
            ('tagid_10', None),  # undefined
            ('thresh_state_green', int),
            ('thresh_state_blue', int),
            ('thresh_state_lo', int),
            ('thresh_state_hi', int),
            ('zoom', int),
            ('create_time', julian_datetime),
            ('last_saved_time', julian_datetime),
            ('current_buffer', int),
            ('gray_fit', None),
            ('gray_point_count', None),
            ('gray_x', Fraction),
            ('gray_y', Fraction),
            ('gray_min', Fraction),
            ('gray_max', Fraction),
            ('gray_unit_name', str),
            ('standard_lut', int),
            ('wavelength', int),
            ('stage_position', '(%i,2,2)u4'),  # N xy positions as fract
            ('camera_chip_offset', '(%i,2,2)u4'),  # N xy offsets as fract
            ('overlay_mask', None),
            ('overlay_compress', None),
            ('overlay', None),
            ('special_overlay_mask', None),
            ('special_overlay_compress', None),
            ('special_overlay', None),
            ('image_property', read_uic_image_property),
            ('stage_label', '%ip'),  # N str
            ('autoscale_lo_info', Fraction),
            ('autoscale_hi_info', Fraction),
            ('absolute_z', '(%i,2)u4'),  # N fractions
            ('absolute_z_valid', '(%i,)u4'),  # N long
            ('gamma', 'I'),  # 'I' uses offset
            ('gamma_red', 'I'),
            ('gamma_green', 'I'),
            ('gamma_blue', 'I'),
            ('camera_bin', '2I'),
            ('new_lut', int),
            ('image_property_ex', None),
            ('plane_property', int),
            ('user_lut_table', '(256,3)u1'),
            ('red_autoscale_info', int),
            ('red_autoscale_lo_info', Fraction),
            ('red_autoscale_hi_info', Fraction),
            ('red_minscale_info', int),
            ('red_maxscale_info', int),
            ('green_autoscale_info', int),
            ('green_autoscale_lo_info', Fraction),
            ('green_autoscale_hi_info', Fraction),
            ('green_minscale_info', int),
            ('green_maxscale_info', int),
            ('blue_autoscale_info', int),
            ('blue_autoscale_lo_info', Fraction),
            ('blue_autoscale_hi_info', Fraction),
            ('blue_min_scale_info', int),
            ('blue_max_scale_info', int),
            # ('overlay_plane_color', read_uic_overlay_plane_color),

        # Bytes with reversed bitorder
        return (
            b'\x00\x80@\xc0 \xa0`\xe0\x10\x90P\xd00\xb0p\xf0\x08\x88H\xc8('

        # Numpy array of bytes with reversed bitorder
        return numpy.fromstring(CONST.REVERSE_BITORDER_BYTES, dtype='uint8')

    # Max line length of printed output

    # Max number of lines to print

def read_bytes(fh, byteorder, dtype, count):
    """Read tag data from file and return as byte string."""
    dtype = 'b' if dtype[-1] == 's' else byteorder+dtype[-1]
    return fh.read_array(dtype, count).tostring()

def read_numpy(fh, byteorder, dtype, count):
    """Read tag data from file and return as numpy array."""
    dtype = 'b' if dtype[-1] == 's' else byteorder+dtype[-1]
    return fh.read_array(dtype, count)

def read_json(fh, byteorder, dtype, count):
    """Read JSON tag data from file and return as object."""
    data = fh.read(count)
        return json.loads(unicode(stripnull(data), 'utf-8'))
    except ValueError:
        warnings.warn("invalid JSON '%s'" % data)

def read_mm_header(fh, byteorder, dtype, count):
    """Read mm_header tag from file and return as numpy.rec.array."""
    return fh.read_record(CONST.FLUOVIEW_MM_HEADER, byteorder=byteorder)

def read_mm_stamp(fh, byteorder, dtype, count):
    """Read MM_STAMP tag from file and return as numpy.ndarray."""
    return fh.read_array(byteorder+'f8', 8)

def read_uic1tag(fh, byteorder, dtype, count, plane_count=None):
    """Read MetaMorph STK UIC1Tag from file and return as dict.

    Return empty dictionary if plane_count is unknown.

    assert dtype in ('2I', '1I') and byteorder == '<'
    result = {}
    if dtype == '2I':
        # pre MetaMorph 2.5 (not tested)
        values = fh.read_array('<u4', 2*count).reshape(count, 2)
        result = {'z_distance': values[:, 0] / values[:, 1]}
    elif plane_count:
        for _ in range(count):
            tagid = struct.unpack('<I', fh.read(4))[0]
            if tagid in (28, 29, 37, 40, 41):
                # silently skip unexpected tags
            name, value = read_uic_tag(fh, tagid, plane_count, offset=True)
            result[name] = value
    return result

def read_uic2tag(fh, byteorder, dtype, plane_count):
    """Read MetaMorph STK UIC2Tag from file and return as dict."""
    assert dtype == '2I' and byteorder == '<'
    values = fh.read_array('<u4', 6*plane_count).reshape(plane_count, 6)
    return {
        'z_distance': values[:, 0] / values[:, 1],
        'date_created': values[:, 2],  # julian days
        'time_created': values[:, 3],  # milliseconds
        'date_modified': values[:, 4],  # julian days
        'time_modified': values[:, 5]}  # milliseconds

def read_uic3tag(fh, byteorder, dtype, plane_count):
    """Read MetaMorph STK UIC3Tag from file and return as dict."""
    assert dtype == '2I' and byteorder == '<'
    values = fh.read_array('<u4', 2*plane_count).reshape(plane_count, 2)
    return {'wavelengths': values[:, 0] / values[:, 1]}

def read_uic4tag(fh, byteorder, dtype, plane_count):
    """Read MetaMorph STK UIC4Tag from file and return as dict."""
    assert dtype == '1I' and byteorder == '<'
    result = {}
    while True:
        tagid = struct.unpack('<H', fh.read(2))[0]
        if tagid == 0:
        name, value = read_uic_tag(fh, tagid, plane_count, offset=False)
        result[name] = value
    return result

def read_uic_tag(fh, tagid, plane_count, offset):
    """Read a single UIC tag value from file and return tag name and value.

    UIC1Tags use an offset.

    def read_int(count=1):
        value = struct.unpack('<%iI' % count, fh.read(4*count))
        return value[0] if count == 1 else value

        name, dtype = CONST.UIC_TAGS[tagid]
    except IndexError:
        # unknown tag
        return '_tagid_%i' % tagid, read_int()

    Fraction = CONST.UIC_TAGS[4][1]

    if offset:
        pos = fh.tell()
        if dtype not in (int, None):
            off = read_int()
            if off < 8:
                if dtype is str:
                    return name, ''
                warnings.warn("invalid offset for uic tag '%s': %i" %
                              (name, off))
                return name, off

    if dtype is None:
        # skip
        name = '_' + name
        value = read_int()
    elif dtype is int:
        # int
        value = read_int()
    elif dtype is Fraction:
        # fraction
        value = read_int(2)
        value = value[0] / value[1]
    elif dtype is julian_datetime:
        # datetime
        value = julian_datetime(*read_int(2))
    elif dtype is read_uic_image_property:
        # ImagePropertyEx
        value = read_uic_image_property(fh)
    elif dtype is str:
        # pascal string
        size = read_int()
        if 0 <= size < 2**10:
            value = struct.unpack('%is' % size, fh.read(size))[0][:-1]
            value = stripnull(value)
        elif offset:
            value = ''
            warnings.warn("corrupt string in uic tag '%s'" % name)
            raise ValueError("invalid string size %i" % size)
    elif dtype == '%ip':
        # sequence of pascal strings
        value = []
        for _ in range(plane_count):
            size = read_int()
            if 0 <= size < 2**10:
                string = struct.unpack('%is' % size, fh.read(size))[0][:-1]
                string = stripnull(string)
            elif offset:
                warnings.warn("corrupt string in uic tag '%s'" % name)
                raise ValueError("invalid string size %i" % size)
        # struct or numpy type
        dtype = '<' + dtype
        if '%i' in dtype:
            dtype = dtype % plane_count
        if '(' in dtype:
            # numpy type
            value = fh.read_array(dtype, 1)[0]
            if value.shape[-1] == 2:
                # assume fractions
                value = value[..., 0] / value[..., 1]
            # struct format
            value = struct.unpack(dtype, fh.read(struct.calcsize(dtype)))
            if len(value) == 1:
                value = value[0]

    if offset:
        fh.seek(pos + 4)

    return name, value

def read_uic_image_property(fh):
    """Read UIC ImagePropertyEx tag from file and return as dict."""
    # TODO: test this
    size = struct.unpack('B', fh.read(1))[0]
    name = struct.unpack('%is' % size, fh.read(size))[0][:-1]
    flags, prop = struct.unpack('<IB', fh.read(5))
    if prop == 1:
        value = struct.unpack('II', fh.read(8))
        value = value[0] / value[1]
        size = struct.unpack('B', fh.read(1))[0]
        value = struct.unpack('%is' % size, fh.read(size))[0]
    return dict(name=name, flags=flags, value=value)

def read_cz_lsm_info(fh, byteorder, dtype, count):
    """Read CS_LSM_INFO tag from file and return as numpy.rec.array."""
    assert byteorder == '<'
    magic_number, structure_size = struct.unpack('<II', fh.read(8))
    if magic_number not in (50350412, 67127628):
        raise ValueError("invalid CS_LSM_INFO structure")
    fh.seek(-8, 1)

    if structure_size < numpy.dtype(CONST.CZ_LSM_INFO).itemsize:
        # adjust structure according to structure_size
        cz_lsm_info = []
        size = 0
        for name, dtype in CONST.CZ_LSM_INFO:
            size += numpy.dtype(dtype).itemsize
            if size > structure_size:
            cz_lsm_info.append((name, dtype))
        cz_lsm_info = CONST.CZ_LSM_INFO

    return fh.read_record(cz_lsm_info, byteorder=byteorder)

def read_cz_lsm_floatpairs(fh):
    """Read LSM sequence of float pairs from file and return as list."""
    size = struct.unpack('<i', fh.read(4))[0]
    return fh.read_array('<2f8', count=size)

def read_cz_lsm_positions(fh):
    """Read LSM positions from file and return as list."""
    size = struct.unpack('<I', fh.read(4))[0]
    return fh.read_array('<2f8', count=size)

def read_cz_lsm_time_stamps(fh):
    """Read LSM time stamps from file and return as list."""
    size, count = struct.unpack('<ii', fh.read(8))
    if size != (8 + 8 * count):
        raise ValueError("lsm_time_stamps block is too short")
    # return struct.unpack('<%dd' % count, fh.read(8*count))
    return fh.read_array('<f8', count=count)

def read_cz_lsm_event_list(fh):
    """Read LSM events from file and return as list of (time, type, text)."""
    count = struct.unpack('<II', fh.read(8))[1]
    events = []
    while count > 0:
        esize, etime, etype = struct.unpack('<IdI', fh.read(16))
        etext = stripnull(fh.read(esize - 16))
        events.append((etime, etype, etext))
        count -= 1
    return events

def read_cz_lsm_scan_info(fh):
    """Read LSM scan information from file and return as dict."""
    block = {}
    blocks = [block]
    unpack = struct.unpack
    if 0x10000000 != struct.unpack('<I', fh.read(4))[0]:
        # not a Recording sub block
        raise ValueError("not a lsm_scan_info structure")
    while True:
        entry, dtype, size = unpack('<III', fh.read(12))
        if dtype == 2:
            # ascii
            value = stripnull(fh.read(size))
        elif dtype == 4:
            # long
            value = unpack('<i', fh.read(4))[0]
        elif dtype == 5:
            # rational
            value = unpack('<d', fh.read(8))[0]
            value = 0
        if entry in CONST.CZ_LSM_SCAN_INFO_ARRAYS:
            name = CONST.CZ_LSM_SCAN_INFO_ARRAYS[entry]
            newobj = []
            block[name] = newobj
            block = newobj
        elif entry in CONST.CZ_LSM_SCAN_INFO_STRUCTS:
            newobj = {}
            block = newobj
        elif entry in CONST.CZ_LSM_SCAN_INFO_ATTRIBUTES:
            name = CONST.CZ_LSM_SCAN_INFO_ATTRIBUTES[entry]
            block[name] = value
        elif entry == 0xffffffff:
            # end sub block
            block = blocks.pop()
            # unknown entry
            block["entry_0x%x" % entry] = value
        if not blocks:
    return block

def read_tvips_header(fh, byteorder, dtype, count):
    """Read TVIPS EM-MENU headers and return as dict."""
    result = {}
    header = fh.read_record(CONST.TVIPS_HEADER_V1, byteorder=byteorder)
    if header.version == 2:
        header = fh.read_record(CONST.TVIPS_HEADER_V2, byteorder=byteorder)
        if header.magic != int(0xaaaaaaaa):
            raise ValueError("invalid TVIPS v2 magic number")
        # decode utf16 strings
        for name, typestr in CONST.TVIPS_HEADER_V2:
            if typestr.startswith('V'):
                s = header[name].tostring().decode('utf16', errors='ignore')
                result[name] = stripnull(s, null='\0')
                result[name] = header[name]
        # convert nm to m
        for axis in 'xy':
            header['physical_pixel_size_' + axis] /= 1e9
            header['pixel_size_' + axis] /= 1e9
    elif header.version == 1:
        for name, typestr in CONST.TVIPS_HEADER_V2:
            result[name] = header[name]
        raise ValueError("unknown TVIPS header version")
    return result

def read_fei_metadata(fh, byteorder, dtype, count):
    """Read FEI SFEG/HELIOS headers and return as nested dict."""
    result = {}
    section = {}
    for line in fh.read(count).splitlines():
        line = line.strip()
        if line.startswith(b'['):
            section = {}
            result[bytes2str(line[1:-1])] = section
            key, value = line.split(b'=')
        except ValueError:
        section[bytes2str(key)] = astype(value)
    return result

def read_sem_metadata(fh, byteorder, dtype, count):
    """Read Zeiss SEM tag and return as dict."""
    result = {'': ()}
    key = None
    for line in fh.read(count).splitlines():
        line = line.decode('cp1252')
        if line.isupper():
            key = line.lower()
        elif key:
                name, value = line.split('=')
            except ValueError:
            value = value.strip()
            unit = ''
                v, u = value.split()
                number = astype(v, (int, float))
                if number != v:
                    value = number
                    unit = u
            except Exception:
                number = astype(value, (int, float))
                if number != value:
                    value = number
                if value in ('No', 'Off'):
                    value = False
                elif value in ('Yes', 'On'):
                    value = True
            result[key] = (name.strip(), value)
            if unit:
                result[key] += (unit,)
            key = None
            result[''] += (astype(line, (int, float)),)
    return result

def read_nih_image_header(fh, byteorder, dtype, count):
    """Read NIH_IMAGE_HEADER tag from file and return as numpy.rec.array."""
    a = fh.read_record(CONST.NIH_IMAGE_HEADER, byteorder=byteorder)
    a = a.newbyteorder(byteorder)
    a.xunit = a.xunit[:a._xunit_len]
    a.um = a.um[:a._um_len]
    return a

def read_scanimage_metadata(fh):
    """Read ScanImage BigTIFF v3 static and ROI metadata from open file.

    Return non-varying frame data as dict and ROI group data as JSON.

    The settings can be used to read image data and metadata without parsing
    the TIFF file.

    Raise ValueError if file does not contain valid ScanImage v3 metadata.

        byteorder, version = struct.unpack('<2sH', fh.read(4))
        if byteorder != b'II' or version != 43:
            raise Exception
        magic, version, size0, size1 = struct.unpack('<IIII', fh.read(16))
        if magic != 117637889 or version != 3:
            raise Exception
    except Exception:
        raise ValueError("not a ScanImage BigTIFF v3 file")

    frame_data = matlabstr2py(bytes2str(fh.read(size0)[:-1]))
    roi_data = read_json(fh, '<', None, size1)
    return frame_data, roi_data

def read_micromanager_metadata(fh):
    """Read MicroManager non-TIFF settings from open file and return as dict.

    The settings can be used to read image data without parsing the TIFF file.

    Raise ValueError if the file does not contain valid MicroManager metadata.

        byteorder = {b'II': '<', b'MM': '>'}[fh.read(2)]