-*- coding: utf-8 -*-

The Python Imaging Library
$Id$

ACKNOWLEDGEMENTS: PIL wouldn't be what it is without the help of:
David Ascher, Phil Austin, Douglas Bagnall, Larry Bates, Anthony
Baxter, William Baxter, Denis Benoit, Jan Blom, Duncan Booth, Alexey
Borzenkov, Jeff Breidenbach, Roger Burnham, Zac Burns, Gene Cash,
Kevin Cazabon, Fred Clare, Greg Coats, Chris Cogdon, Greg Couch, Bill
Crutchfield, Abel Deuring, Tim Docker, Fred Drake, Graham Dumpleton,
Matthew Ellis, Eric Etheridge, Daniel Fetchinson, Robin Friedrich,
Pier Paolo Glave, Federico Di Gregorio, Markus Gritsch, Daniel
Haertle, Greg Hamilton, Mark Hammond, Bernhard Herzog, Rob Hooft, Bob
Ippolito, Jack Jansen, Bill Janssen, Edward Jones, Richard Jones,
Håkan Karlsson, Robert Kern, David Kirtley, Bob Klimek, Matthias
Klose, Andrew Kuchling, Magnus Källström, Victor Lacina, Ben Last,
Hamish Lawson, Cesare Leonardi, Andrew MacIntyre, Jan Matejek, Naveen
Michaud-Agrawal, Gordon McMillan, Skip Montanaro, Fredrik Nehr,
Russell Nelson, Luciano Nocera, Travis Oliphant, Piet van Oostrum,
Richard Oudkerk, Paul Pharr, Andres Polit, Conrado Porto Lopes Gouvêa,
Eric Raymond, Victor Reijs, Bertil Reinhammar, Nicholas Riley, Don
Rozenberg, Toby Sargeant, Barry Scott, Les Schaffer, Joel Shprentz,
Klamer Shutte, Gene Skonicki, Niki Spahiev, D. Alan Stewart, Perry
Stoll, Paul Svensson, Ulrik Svensson, Miki Tebeka, Michael van
Tellingen, Ivan Tkatchev, Dan Torop, Adam Twardoch, Rune Uhlin, Dmitry
Vasiliev, Sasha Voynow, Charles Waldman, Collin Winter, Dan Wolfe,
Ka-Ping Yee, and many others (if your name should be on this list, let
me know.)

*** Changes from release 1.1.6 to 1.1.7 ***

This section may not be fully complete.  For changes since this file
was last updated, see the repository revision history:

  http://bitbucket.org/effbot/pil-2009-raclette/changesets/

(1.1.7 final)

+ Set GIF loop info property to the number of iterations if a NETSCAPE
  loop extension is present, instead of always setting it to 1 (from
  Valentino Volonghi).

(1.1.7c1 released)

+ Improved PNG compression (from Alexey Borzenkov).

+ Read interlaced PNG files (from Conrado Porto Lopes Gouvêa)

+ Added various TGA improvements from Alexey Borzenkov, including
  support for specifying image orientation.

+ Bumped block threshold to 16 megabytes, made size estimation a bit
  more accurate.  This speeds up allocation of large images.

+ Fixed rounding error in ImagingDrawWideLine.

  "gormish" writes: ImagingDrawWideLine() in Draw.c has a bug in every
  version I've seen, which leads to different width lines depending on
  the order of the points in the line. This is especially bad at some
  angles where a 'width=2' line can completely disappear.

+ Added support for RGBA mode to the SGI module (based on code by
  Karsten Hiddemann).

+ Handle repeated IPTC tags (adapted from a patch by Eric Bruning).

  Eric writes: According to the specification, some IPTC tags can be
  repeated, e.g., tag 2:25 (keywords). PIL 1.1.6 only retained the last
  instance of that tag. Below is a patch to store all tags. If there are
  multiple tag instances, they are stored in a (python) list. Single tag
  instances remain as strings.

+ Fixed potential crash in ImageFilter for small target images
  (reported by Zac Burns and Daniel Fetchinson).

+ Use BMP instead of JPEG as temporary show format on Mac OS X.

+ Fixed putpixel/new for I;16 with colors > 255.

+ Added integer power support to ImagingMath.

+ Added limited support for I;16L mode (explicit little endian).

+ Moved WMF support into Image.core; enable WMF rendering by default
  if renderer is available.

+ Mark the ARG plugin as obsolete.

+ Added version query mechanism to ImageCms and ImageFont, for
  debugging.

+ Added (experimental) ImageCms function for fetching the ICC profile
  for the current display (currently Windows only).

  Added HWND/HDC support to ImageCms.get_display_profile().

+ Added WMF renderer (Windows only).

+ Added ImagePointHandler and ImageTransformHandler mixins; made
  ImageCmsTransform work with im.point.

+ Fixed potential endless loop in the XVThumbnail reader (from Nikolai
  Ugelvik).

+ Added Kevin Cazabon's pyCMS package.

  The C code has been moved to _imagingcms.c, the Python interface
  module is installed as PIL.ImageCMS.

  Added support for in-memory ICC profiles.

  Unified buildTransform and buildTransformFromOpenProfiles.

  The profile can now be either a filename, a profile object, or a
  file-like object containing an in-memory profile.

  Additional fixes from Florian Böch:

    Very nice - it just needs LCMS flags support so we can use black
    point compensation and softproofing :) See attached patches.  They
    also fix a naming issue which could cause confusion - display
    profile (ImageCms wording) actually means proof profile (lcms
    wording), so I changed variable names and docstrings where
    applicable. Patches are tested under Python 2.6.

+ Improved support for layer names in PSD files (from Sylvain Baubeau)

  Sylvain writes: I needed to be able to retrieve the names of the
  layers in a PSD files. But PsdImagePlugin.py didn't do the job so I
  wrote this very small patch.

+ Improved RGBA support for ImageTk for 8.4 and newer (from Con
  Radchenko).

  This replaces the slow run-length based encoding model with true
  compositing at the Tk level.

+ Added support for 16- and 32-bit images to McIdas loader.

  Based on file samples and stand-alone reader code provided by Craig
  Swank.

+ Added ImagePalette support to putpalette.

+ Fixed problem with incremental parsing of PNG files.

+ Make selftest.py report non-zero status on failure (from Mark
  Sienkiewicz)

+ Add big endian save support and multipage infrastructure to the TIFF
  writer (from Sebastian Haase).

+ Handle files with GPS IFD but no basic EXIF IFD (reported by Kurt
  Schwehr).

+ Added zTXT support (from Andrew Kuchling via Lowell Alleman).

+ Fixed potential infinite loop bug in ImageFont (from Guilherme Polo).

+ Added sample ICC profiles (from Kevin Cazabon)

+ Fixed array interface for I, F, and RGBA/RGBX images.

+ Added Chroma subsampling support for JPEG (from Justin Huff).

  Justin writes: Attached is a patch (against PIL 1.1.6) to provide
  control over the chroma subsampling done by the JPEG encoder.  This
  is often useful for reducing compression artifacts around edges of
  clipart and text.

+ Added USM/Gaussian Blur code from Kevin Cazabon.

+ Fixed bug w. uninitialized image data when cropping outside the
  source image.

+ Use ImageShow to implement the Image.show method.

  Most notably, this picks the 'display' utility when available.  It
  also allows application code to register new display utilities via
  the ImageShow registry.

+ Release the GIL in the PNG compressor (from Michael van Tellingen).

+ Revised JPEG CMYK handling.

  Always assume Adobe behaviour, both when reading and writing (based on
  a patch by Kevin Cazabon, and test data by Tim V. and Charlie Clark, and
  additional debugging by Michael van Tellingen).

+ Support for preserving ICC profiles (by Florian Böch via Tim Hatch).

  Florian writes:

  It's a beta, so still needs some testing, but should allow you to:
  - retain embedded ICC profiles when saving from/to JPEG, PNG, TIFF.
     Existing code doesn't need to be changed.
  - access embedded profiles in JPEG, PNG, PSD, TIFF.

  It also includes patches for TIFF to retain IPTC, Photoshop and XMP
  metadata when saving as TIFF again, read/write TIFF resolution
  information correctly, and to correct inverted CMYK JPEG files.

+ Fixed potential memory leak in median cut quantizer (from Evgeny Salmin).

+ Fixed OverflowError when reading upside-down BMP images.

+ Added resolution save option for PDF files.

  Andreas Kostyrka writes: I've included a patched PdfImagePlugin.py
  based on 1.1.6 as included in Ubuntu, that supports a "resolution"
  save option. Not great, but it makes the PDF saving more useful by
  allowing PDFs that are not exactly 72dpi.

+ Look for Tcl/Tk include files in version-specific include directory
  (from Encolpe Degoute).

+ Fixed grayscale rounding error in ImageColor.getcolor (from Tim
  Hatch).

+ Fixed calculation of mean value in ImageEnhance.Contrast (reported
  by "roop" and Scott David Daniels).

+ Fixed truetype positioning when first character has a negative left
  bearing (from Ned Batchelder):

  Ned writes: In PIL 1.1.6, ImageDraw.text will position the string
  incorrectly if the first character has a negative left bearing.  To
  see the problem, show a string like "///" in an italic font.  The
  first slash will be clipped at the left, and the string will be
  mis-positioned.

+ Fixed resolution unit bug in tiff reader/writer (based on code by
  Florian Höch, Gary Bloom, and others).

+ Added simple transparency support for RGB images (reported by
  Sebastian Spaeth).

+ Added support for Unicode filenames in ImageFont.truetype (from Donn
  Ingle).

+ Fixed potential crash in ImageFont.getname method (from Donn Ingle).

+ Fixed encoding issue in PIL/WalImageFile (from Santiago M. Mola).

*** Changes from release 1.1.5 to 1.1.6 ***

(1.1.6 released)

+ Fixed some 64-bit compatibility warnings for Python 2.5.

+ Added threading support for the Sane driver (from Abel Deuring).

(1.1.6b2 released)

+ Added experimental "floodfill" function to the ImageDraw module
  (based on code by Eric Raymond).

+ The default arguments for "frombuffer" doesn't match "fromstring"
  and the documentation; this is a bug, and will most likely be fixed
  in a future version.  In this release, PIL prints a warning message
  instead.  To silence the warning, change any calls of the form
  "frombuffer(mode, size, data)" to

      frombuffer(mode, size, data, "raw", mode, 0, 1)

+ Added "fromarray" function, which takes an object implementing the
  NumPy array interface and creates a PIL Image from it. (from Travis
  Oliphant).

+ Added NumPy array interface support (__array_interface__) to the
  Image class (based on code by Travis Oliphant).

  This allows you to easily convert between PIL image memories and
  NumPy arrays:

	import numpy, Image

	im = Image.open('lena.jpg')

	a = numpy.asarray(im) # a is readonly

	im = Image.fromarray(a)

+ Fixed CMYK polarity for JPEG images, by treating all images as
  "Adobe CMYK" images. (thanks to Cesare Leonardi and Kevin Cazabon
  for samples, debugging, and patches).

(1.1.6b1 released)

+ Added 'expand' option to the Image 'rotate' method.  If true, the
  output image is made large enough to hold the entire rotated image.

+ Changed the ImageDraw 'line' method to always draw the last pixel in
  a polyline, independent of line angle.

+ Fixed bearing calculation and clipping in the ImageFont truetype
  renderer; this could lead to clipped text, or crashes in the low-
  level _imagingft module.  (based on input from Adam Twardoch and
  others).

+ Added ImageQt wrapper module, for converting PIL Image objects to
  QImage objects in an efficient way.

+ Fixed 'getmodebands' to return the number of bands also for "PA"
  and "LA" modes.  Added 'getmodebandnames' helper that return the
  band names.

(1.1.6a2 released)

+ Added float/double support to the TIFF loader (from Russell
  Nelson).

+ Fixed broken use of realloc() in path.c (from Jan Matejek)

+ Added save support for Spider images (from William Baxter).

+ Fixed broken 'paste' and 'resize' operations in pildriver
  (from Bill Janssen).

+ Added support for duplex scanning to the Sane interface (Abel
  Deuring).

(1.1.6a1 released)

+ Fixed a memory leak in "convert(mode)", when converting from
  L to P.

+ Added pixel access object.  The "load" method now returns a
  access object that can be used to directly get and set pixel
  values, using ordinary [x, y] notation:

	pixel = im.load()
	v = pixel[x, y]	
	pixel[x, y] = v	

  If you're accessing more than a few pixels, this is a lot
  faster than using getpixel/putpixel.

+ Fixed building on Cygwin (from Miki Tebeka).

+ Fixed "point(callable)" on unloaded images (reported by Håkan
  Karlsson).

+ Fixed size bug in ImageWin.ImageWindow constructor (from Victor
  Reijs)

+ Fixed ImageMath float() and int() operations for Python 2.4
  (reported by Don Rozenberg).

+ Fixed "RuntimeError: encoder error -8 in tostring" problem for
  wide "RGB", "I", and "F" images.

+ Fixed line width calculation.

(1.1.6a0 released)

+ Fixed byte order issue in Image.paste(ink) (from Ka-Ping Yee).

+ Fixed off-by-0.5 errors in the ANTIALIAS code (based on input
  from Douglas Bagnall).

+ Added buffer interface support to the Path constructor.  If
  a buffer is provided, it is assumed to contain a flat array
  of float coordinates (e.g. array.array('f', seq)).

+ Added new ImageMath module.

+ Fixed ImageOps.equalize when used with a small number of distinct
  values (reported by David Kirtley).

+ Fixed potential integer division in PSDraw.image (from Eric Etheridge).

*** Changes from release 1.1 to 1.1.5 ***

(1.1.5c2 and 1.1.5 final released)

+ Added experimental PERSPECTIVE transform method (from Jeff Breiden-
  bach).

(1.1.5c1 released)

+ Make sure "thumbnail" never generates zero-wide or zero-high images
  (reported by Gene Skonicki)

+ Fixed a "getcolors" bug that could result in a zero count for some
  colors (reported by Richard Oudkerk).

+ Changed default "convert" palette to avoid "rounding errors" when
  round-tripping white source pixels (reported by Henryk Gerlach and
  Jeff Epler).

(1.1.5b3 released)

+ Don't crash in "quantize" method if the number of colors requested
  is larger than 256.  This release raises a ValueError exception;
  future versions may return a mode "RGB" image instead (reported
  by Richard Oudkerk).

+ Added WBMP read/write support (based on code by Duncan Booth).

(1.1.5b2 released)

+ Added DPI read/write support to the PNG codec.  The decoder sets
  the info["dpi"] attribute for PNG files with appropriate resolution
  settings.  The encoder uses the "dpi" option (based on code by Niki
  Spahiev).

+ Added limited support for "point" mappings from mode "I" to mode "L".
  Only 16-bit values are supported (other values are clipped), the lookup
  table must contain exactly 65536 entries, and the mode argument must be
  set to "L".

+ Added support for Mac OS X icns files (based on code by Bob Ippolito).

+ Added "ModeFilter" support to the ImageFilter module.

+ Added support for Spider images (from William Baxter).  See the
  comments in PIL/SpiderImagePlugin.py for more information on this
  format.

(1.1.5b1 released)

+ Added new Sane release (from Ralph Heinkel).  See the Sane/README
  and Sane/CHANGES files for more information.

+ Added experimental PngInfo chunk container to the PngImageFile
  module.  This can be used to add arbitrary chunks to a PNG file.
  Create a PngInfo instance, use "add" or "add_text" to add chunks,
  and pass the instance as the "pnginfo" option when saving the
  file.

+ Added "getpalette" method.  This returns the palette as a list,
  or None if the image has no palette.  To modify the palette, use
  "getpalette" to fetch the current palette, modify the list, and
  put it back using "putpalette".

+ Added optional flattening to the ImagePath "tolist" method.
  tolist() or tolist(0) returns a list of 2-tuples, as before.
  tolist(1) returns a flattened list instead.

(1.1.5a5 released)

+ Fixed BILINEAR/BICUBIC/ANTIALIAS filtering for mode "LA".

+ Added "getcolors()" method.  This is similar to the existing histo-
  gram method, but looks at color values instead of individual layers,
  and returns an unsorted list of (count, color) tuples.

  By default, the method returns None if finds more than 256 colors.
  If you need to look for more colors, you can pass in a limit (this
  is used to allocate internal tables, so you probably don't want to
  pass in too large values).

+ Build improvements: Fixed building under AIX, improved detection of
  FreeType2 and Mac OS X framework libraries, and more.  Many thanks
  to everyone who helped test the new "setup.py" script!

(1.1.5a4 released)

+ The "save" method now looks for a file format driver before
  creating the file.

+ Don't use antialiased truetype fonts when drawing in mode "P", "I",
  and "F" images.

+ Rewrote the "setup.py" file.  The new version scans for available
  support libraries, and configures both the libImaging core library
  and the bindings in one step.

  To use specific versions of the libraries, edit the ROOT variables
  in the setup.py file.

+ Removed threaded "show" viewer; use the old "show" implementation
  instead (Windows).

+ Added deprecation warnings to Image.offset, ImageDraw.setink, and
  ImageDraw.setfill.

+ Added width option to ImageDraw.line().  The current implementation
  works best for straight lines; it does not support line joins, so
  polylines won't look good.

+ ImageDraw.Draw is now a factory function instead of a class.  If
  you need to create custom draw classes, inherit from the ImageDraw
  class.    All other code should use the factory function.

+ Fixed loading of certain PCX files (problem reported by Greg
  Hamilton, who also provided samples).

+ Changed _imagingft.c to require FreeType 2.1 or newer.  The
  module can still be built with earlier versions; see comments
  in _imagingft.c for details.

(1.1.5a3 released)

+ Added 'getim' method, which returns a PyCObject wrapping an
  Imaging pointer.  The description string is set to IMAGING_MAGIC.
  See Imaging.h for pointer and string declarations.

+ Fixed reading of TIFF JPEG images (problem reported by Ulrik
  Svensson).

+ Made ImageColor work under Python 1.5.2

+ Fixed division by zero "equalize" on very small images (from
  Douglas Bagnall).

(1.1.5a2 released)

+ The "paste" method now supports the alternative "paste(im, mask)"
  syntax (in this case, the box defaults to im's bounding box).

+ The "ImageFile.Parser" class now works also for PNG files with
  more than one IDAT block.

+ Added DPI read/write to the TIFF codec, and fixed writing of
  rational values.  The decoder sets the info["dpi"] attribute
  for TIFF files with appropriate resolution settings.  The
  encoder uses the "dpi" option.

+ Disable interlacing for small (or narrow) GIF images, to
  work around what appears to be a hard-to-find bug in PIL's
  GIF encoder.

+ Fixed writing of mode "P" PDF images.  Made mode "1" PDF
  images smaller.

+ Made the XBM reader a bit more robust; the file may now start
  with a few whitespace characters.

+ Added support for enhanced metafiles to the WMF driver.  The
  separate PILWMF kit lets you render both placeable WMF files
  and EMF files as raster images.  See

      http://effbot.org/downloads#pilwmf

(1.1.5a1 released)

+ Replaced broken WMF driver with a WMF stub plugin (see below).

+ Fixed writing of mode "1", "L", and "CMYK" PDF images (based on
  input from Nicholas Riley and others).

+ Fixed adaptive palette conversion for zero-width or zero-height
  images (from Chris Cogdon)

+ Fixed reading of PNG images from QuickTime 6 (from Paul Pharr)

+ Added support for StubImageFile plugins, including stub plugins
  for BUFR, FITS, GRIB, and HDF5 files.  A stub plugin can identify
  a given file format, but relies on application code to open and
  save files in that format.

+ Added optional "encoding" argument to the ImageFont.truetype
  factory.  This argument can be used to specify non-Unicode character
  maps for fonts that support that.  For example, to draw text using
  the Microsoft Symbol font, use:

      font = ImageFont.truetype("symbol.ttf", 16, encoding="symb")
      draw.text((0, 0), unichr(0xF000 + 0xAA))

  (note that the symbol font uses characters in the 0xF000-0xF0FF
   range)

  Common encodings are "unic" (Unicode), "symb" (Microsoft Symbol),
  "ADOB" (Adobe Standard), "ADBE" (Adobe Expert), and "armn" (Apple
  Roman).  See the FreeType documentation for more information.

+ Made "putalpha" a bit more robust; you can now attach an alpha
  layer to a plain "L" or "RGB" image, and you can also specify
  constant alphas instead of alpha layers (using integers or colour
  names).

+ Added experimental "LA" mode support.

  An "LA" image is an "L" image with an attached transparency layer.
  Note that support for "LA" is not complete; some operations may
  fail or produce unexpected results.

+ Added "RankFilter", "MinFilter", "MedianFilter", and "MaxFilter"
  classes to the ImageFilter module.

+ Improved support for applications using multiple threads; PIL
  now releases the global interpreter lock for many CPU-intensive
  operations (based on work by Kevin Cazabon).

+ Ignore Unicode characters in the PCF loader (from Andres Polit)

+ Fixed typo in OleFileIO.loadfat, which could affect loading of
  FlashPix and Image Composer images (Daniel Haertle)

+ Fixed building on platforms that have Freetype but don't have
  Tcl/Tk (Jack Jansen, Luciano Nocera, Piet van Oostrum and others)

+ Added EXIF GPSInfo read support for JPEG files.  To extract
  GPSInfo information, open the file, extract the exif dictionary,
  and check for the key 0x8825 (GPSInfo).  If present, it contains
  a dictionary mapping GPS keys to GPS values.  For a list of keys,
  see the EXIF specification.

  The "ExifTags" module contains a GPSTAGS dictionary mapping GPS
  tags to tag names.

+ Added DPI read support to the PCX and DCX codecs (info["dpi"]).

+ The "show" methods now uses a built-in image viewer on Windows.
  This viewer creates an instance of the ImageWindow class (see
  below) and keeps it running in a separate thread.  NOTE: This
  was disabled in 1.1.5a4.

+ Added experimental "Window" and "ImageWindow" classes to the
  ImageWin module.  These classes allow you to create a WCK-style
  toplevel window, and use it to display raster data.

+ Fixed some Python 1.5.2 issues (to build under 1.5.2, use the
  Makefile.pre.in/Setup.in approach)

+ Added support for the TIFF FillOrder tag.  PIL can read mode "1",
  "L", "P" and "RGB" images with non-standard FillOrder (based on
  input from Jeff Breidenbach).

(1.1.4 final released)

+ Fixed ImageTk build problem on Unix.

(1.1.4b2 released)

+ Improved building on Mac OS X (from Jack Jansen).

+ Improved building on Windows with MinGW (from Klamer Shutte).

+ If no font is specified, ImageDraw now uses the embedded default
  font.  Use the "load" or "truetype" methods to load a real font.

+ Added embedded default font to the ImageFont module (currently
  an 8-pixel Courier font, taken from the X window distribution).

(1.1.4b1 released)

+ Added experimental EXIF support for JPEG files.  To extract EXIF
  information from a JPEG file, open the file as usual, and call the
  "_getexif" method.  If successful, this method returns a dictionary
  mapping EXIF TIFF tags to values.  If the file does not contain EXIF
  data, the "_getexif" method returns None.

  The "ExifTags" module contains a dictionary mapping tags to tag
  names.

  This interface will most likely change in future versions.

+ Fixed a bug when using the "transparency" option with the GIF
  writer.

+ Added limited support for "bitfield compression" in BMP files
  and DIB buffers, for 15-bit, 16-bit, and 32-bit images.  This
  also fixes a problem with ImageGrab module when copying screen-
  dumps from the clipboard on 15/16/32-bit displays.

+ Added experimental WAL (Quake 2 textures) loader.  To use this
  loader, import WalImageFile and call the "open" method in that
  module.

(1.1.4a4 released)

+ Added updated SANE driver (Andrew Kuchling, Abel Deuring)

+ Use Python's "mmap" module on non-Windows platforms to read some
  uncompressed formats using memory mapping.  Also added a "frombuffer"
  function that allows you to access the contents of an existing string
  or buffer object as if it were an image object.

+ Fixed a memory leak that could appear when processing mode "P"
  images (from Pier Paolo Glave)

+ Ignore Unicode characters in the BDF loader (from Graham Dumpleton)

(1.1.4a3 released; windows only)

+ Added experimental RGBA-on-RGB drawing support.  To use RGBA
  colours on an RGB image, pass "RGBA" as the second string to
  the ImageDraw.Draw constructor.

+ Added support for non-ASCII strings (Latin-1) and Unicode
  to the truetype font renderer.

+ The ImageWin "Dib" object can now be constructed directly from
  an image object.

+ The ImageWin module now allows you use window handles as well
  as device contexts.  To use a window handle, wrap the handle in
  an ImageWin.HWND object, and pass in this object instead of the
  device context.

(1.1.4a2 released)

+ Improved support for 16-bit unsigned integer images (mode "I;16").
  This includes TIFF reader support, and support for "getextrema"
  and "point" (from Klamer Shutte).

+ Made the BdfFontFile reader a bit more robust (from Kevin Cazabon
  and Dmitry Vasiliev)

+ Changed TIFF writer to always write Compression tag, even when
  using the default compression (from Greg Couch).

+ Added "show" support for Mac OS X (from Dan Wolfe).

+ Added clipboard support to the "ImageGrab" module (Windows only).
  The "grabclipboard" function returns an Image object, a list of
  filenames (not in 1.1.4), or None if neither was found.

(1.1.4a1 released)

+ Improved support for drawing RGB data in palette images.  You can
  now use RGB tuples or colour names (see below) when drawing in a
  mode "P" image.  The drawing layer automatically assigns color
  indexes, as long as you don't use more than 256 unique colours.

+ Moved self test from MiniTest/test.py to ./selftest.py.

+ Added support for CSS3-style color strings to most places that
  accept colour codes/tuples.  This includes the "ImageDraw" module,
  the Image "new" function, and the Image "paste" method.

  Colour strings can use one of the following formats: "#f00",
  "#ff0000", "rgb(255,0,0)", "rgb(100%,0%,0%)", "hsl(0, 100%, 50%)",
  or "red" (most X11-style colour names are supported).  See the
  documentation for the "ImageColor" module for more information.

+ Fixed DCX decoder (based on input from Larry Bates)

+ Added "IptcImagePlugin.getiptcinfo" helper to extract IPTC/NAA
  newsphoto properties from JPEG, TIFF, or IPTC files.

+ Support for TrueType/OpenType fonts has been added to
  the standard distribution.  You need the freetype 2.0
  library.

+ Made the PCX reader a bit more robust when reading 2-bit
  and 4-bit PCX images with odd image sizes.

+ Added "Kernel" class to the ImageFilter module.  This class
  allows you to filter images with user-defined 3x3 and 5x5
  convolution kernels.

+ Added "putdata" support for mode "I", "F" and "RGB".

+ The GIF writer now supports the transparency option (from
  Denis Benoit).

+ A HTML version of the module documentation is now shipped
  with the source code distribution.  You'll find the files in
  the Doc subdirectory.

+ Added support for Palm pixmaps (from Bill Janssen).  This
  change was listed for 1.1.3, but the "PalmImagePlugin" driver
  didn't make it into the distribution.

+ Improved decoder error messages.

(1.1.3 final released)

+ Made setup.py look for old versions of zlib.  For some back-
  ground, see: http://www.gzip.org/zlib/advisory-2002-03-11.txt

(1.1.3c2 released)

+ Added setup.py file (tested on Unix and Windows).  You still
  need to build libImaging/imaging.lib in the traditional way,
  but the setup.py script takes care of the rest.

  The old Setup.in/Makefile.pre.in build method is still
  supported.

+ Fixed segmentation violation in ANTIALIAS filter (an internal
  buffer wasn't properly allocated).

(1.1.3c1 released)

+ Added ANTIALIAS downsampling filter for high-quality "resize"
  and "thumbnail" operations.  Also added filter option to the
  "thumbnail" operation; the default value is NEAREST, but this
  will most likely change in future versions.

+ Fixed plugin loader to be more robust if the __file__
  variable isn't set.

+ Added seek/tell support (for layers) to the PhotoShop
  loader.  Layer 0 is the main image.

+ Added new (but experimental) "ImageOps" module, which provides
  shortcuts for commonly used operations on entire images.

+ Don't mess up when loading PNG images if the decoder leaves
  data in the output buffer.  This could cause internal errors
  on some PNG images, with some versions of ZLIB. (Bug report
  and patch provided by Bernhard Herzog.)

+ Don't mess up on Unicode filenames.

+ Don't mess up when drawing on big endian platforms.

+ Made the TIFF loader a bit more robust; it can now read some
  more slightly broken TIFF files (based on input from Ted Wright,
  Bob Klimek, and D. Alan Stewart)

+ Added OS/2 EMX build files (from Andrew MacIntyre)

+ Change "ImageFont" to reject image files if they don't have the
  right mode.  Older versions could leak memory for "P" images.
  (Bug reported by Markus Gritsch).
  
+ Renamed some internal functions to avoid potential build
  problem on Mac OS X.

+ Added DL_EXPORT where relevant (for Cygwin, based on input
  from Robert Yodlowski)

+ (re)moved bogus __init__ call in BdfFontFile (bug spotted
  by Fred Clare)

+ Added "ImageGrab" support (Windows only)

+ Added support for XBM hotspots (based on code contributed by
  Bernhard Herzog).

+ Added write support for more TIFF tags, namely the Artist,
  Copyright, DateTime, ResolutionUnit, Software, XResolution and
  YResolution tags (from Greg Couch)

+ Added TransposedFont wrapper to ImageFont module

+ Added "optimize" flag to GIF encoder.  If optimize is present
  and non-zero, PIL will work harder to create a small file.

+ Raise "EOFError" (not IndexError) when reading beyond the
  end of a TIFF sequence.

+ Support rewind ("seek(0)") for GIF and TIFF sequences.

+ Load grayscale GIF images as mode "L"

+ Added DPI read/write support to the JPEG codec.  The decoder
  sets the info["dpi"] attribute for JPEG files with JFIF dpi
  settings.  The encoder uses the "dpi" option:

      im = Image.open("file.jpg")
      dpi = im.info["dpi"] # raises KeyError if DPI not known
      im.save("out.jpg", dpi=dpi)

  Note that PIL doesn't always preserve the "info" attribute
  for normal image operations.

(1.1.2c1 and 1.1.2 final released)

+ Adapted to Python 2.1.  Among other things, all uses of the
  "regex" module has been repleased with "re".

+ Fixed attribute error when reading large PNG files (this bug
  was introduced in maintenance code released after the 1.1.1
  release)

+ Ignore non-string objects in sys.path

+ Fixed Image.transform(EXTENT) for negative xoffsets

+ Fixed loading of image plugins if PIL is installed as a package.
  (The plugin loader now always looks in the directory where the
  Image.py module itself is found, even if that directory isn't on
  the standard search path)

+ The Png plugin has been added to the list of preloaded standard
  formats

+ Fixed bitmap/text drawing in fill mode.

+ Fixed "getextrema" to work also for multiband images.

+ Added transparency support for L and P images to the PNG codec.

+ Improved support for read-only images.  The "load" method now
  sets the "readonly" attribute for memory-mapped images.  Operations
  that modifies an image in place (such as "paste" and drawing operations)
  creates an in-memory copy of the image, if necessary.  (before this
  change, any attempt to modify a memory-mapped image resulted in a
  core dump...)

+ Added special cases for lists everywhere PIL expects a sequence.
  This should speed up things like "putdata" and drawing operations.

+ The Image.offset method is deprecated.  Use the ImageChops.offset
  function instead. 

+ Changed ImageChops operators to copy palette and info dictionary
  from the first image argument.

(1.1.1 released)

+ Additional fixes for Python 1.6/2.0, including TIFF "save" bug. 

+ Changed "init" to properly load plugins when PIL is used as a
  package.

+ Fixed broken "show" method (on Unix)

*** Changes from release 1.0 to 1.1 ***

+ Adapted to Python 1.6 ("append" and other method changes)

+ Fixed Image.paste when pasting with solid colour and matte
  layers ("L" or "RGBA" masks) (bug reported by Robert Kern)

+ To make it easier to distribute prebuilt versions of PIL,
  the tkinit binding stuff has been moved to a separate
  extension module, named "_imagingtk".

*** Changes from release 0.3b2 to 1.0 final ***

+ If there's no 16-bit integer (like on a Cray T3E), set
  INT16 to the smallest integer available.  Most of the
  library works just fine anyway (from Bill Crutchfield)

+ Tweaks to make drawing work on big-endian platforms.

(1.0c2 released)

+ If PIL is built with the WITH_TKINTER flag, ImageTk can
  automatically hook into a standard Tkinter build.  You
  no longer need to build your own Tkinter to use the
  ImageTk module.

  The old way still works, though.  For more information,
  see Tk/install.txt.

+ Some tweaks to ImageTk to support multiple Tk interpreters
  (from Greg Couch).

+ ImageFont "load_path" now scans directory mentioned in .pth
  files (from Richard Jones).

(1.0c1 released)

+ The TIFF plugin has been rewritten.  The new plugin fully
  supports all major PIL image modes (including F and I).

+ The ImageFile module now includes a Parser class, which can
  be used to incrementally decode an image file (while down-
  loading it from the net, for example).  See the handbook for
  details.

+ "show" now converts non-standard modes to "L" or "RGB" (as
  appropriate), rather than writing weird things to disk for
  "xv" to choke upon. (bug reported by Les Schaffer).

(1.0b2 released)

+ Major speedups for rotate, transform(EXTENT), and transform(AFFINE)
  when using nearest neighbour resampling.

+ Modified ImageDraw to be compatible with the Arrow graphics
  interface.  See the handbook for details.

+ PIL now automatically loads file codecs when used as a package
  (from The Dragon De Monsyne).  Also included an __init__.py file
  in the standard distribution.

+ The GIF encoder has been modified to produce much smaller files.

  PIL now uses a run-length encoding method to encode GIF files.
  On a random selection of GIF images grabbed from the web, this
  version makes the images about twice as large as the original
  LZW files, where the earlier version made them over 5 times
  larger.  YMMV, of course.

+ Added PCX write support (works with "1", "P", "L", and "RGB")

+ Added "bitmap" and "textsize" methods to ImageDraw.

+ Improved font rendering code.  Fixed a bug or two, and moved
  most of the time critical stuff to C.

+ Removed "bdf2pil.py".  Use "pilfont.py" instead!

+ Improved 16-bit support (still experimental, though).

  The following methods now support "I;16" and "I;16B" images:
  "getpixel", "copy", "convert" (to and from mode "I"), "resize",
  "rotate", and "transform" with nearest neighbour filters, and
  "save" using the IM format.  The "new" and "open" functions
  also work as expected.  On Windows, 16-bit files are memory
  mapped.

  NOTE: ALL other operations are still UNDEFINED on 16-bit images.

+ The "paste" method now supports constant sources.

  Just pass a colour value (a number or a tuple, depending on
  the target image mode) instead of the source image.

