dirac_0.5.3
===========
- Fixed encoder CLI. Added rigorous command line input checks . 
  The encoder exits with error message when command line argument is not 
  valid.
- Fixed HD block size to improve motion compensation and prediction
- Fixed bug with computing MV costs.
- MMX opts to wavelet transform, motion compensation and upconversion which
  improved decoder speed by 25-30%.
- Improved chroma fidelity by increasing bits allocated to chroma
- Wavelet filter type is part of bitstream.
- Fixed bug with the way median is calculated in motion prediction.
- Replaced constant quality encoding control mechanism with a mechanism
  that used constant constant Lagrangian paramemters. 
- MMX optimisations are automatically used if underlying architecture
  supports it.
- Improved motion estimation.
- Modified arithmetic coding to use a look-up table for division.

- Patches: 1211899, 1216297
- Bugs: 1209017, 1209053, 1212581, 1245129, 1240709

dirac_0.5.2
===========
  - Changed encoder command line interface to accept all encoding parameters.
    E.g. frame dimensions, chroma format, frame rate, etc. Header files
    are no longer required. Refer to section 4.3 of README for details.
  - Applied patch supplied by Mike Ferenduros to make the TwoDArray class
    more efficient
  - An overall increase in decoding speed by 20% by speeding up Motion 
    compensation, frame clip, and wavelet transform.
  - Arithmetic coding speedup
  - Added support for multiple wavelet filter types
  - Added support for variable motion vector precision.
  - Updated bitstream version. Encoder/Decoder apps in this release will not
    be able to decode bitstreams generated by earlier versions.
  - Released a source code patch for MPlayer-1.0pre7 to enable Dirac play
    back support.
  - Released a source code patch for ffmpeg-0.4.9-pre1 to enable Dirac encode
    and decode.
  - Released a source code patch for transcode-0.6.14 to enable Dirac encode
    and decode using the patched ffmpeg-0.4.9-pre1.
  - Patches Applied: 1151976 (partial), 1156674
  - Bugs fixed : 1159031

dirac_0.5.1
===========
  - Significant speedup in the Arithmetic coding and decoding engines
  - Modified wavelet coefficient coding to code blocks of coeffients with
    skip flags if all coefficients are zero. This improves compression
    performance and decoding speed.
  - Modified block prediction mode coding to improve compression performance.
  - Fixed bug in assigning costs to lists of vectors in block matching.
  - MMX optimisation for calculating block differences in Motion Estimation.
    This resulted in 20-30% improvement in encoding speed. The configure
    flag --enable-mmx enables compile time MMX optimisation detection.
    Currently works only on GNU/Linux.
  - New script create_dirac_testfile.pl to generate input formats 
    compatible with Dirac from rgb files. Step 3, Section 4.2 in README 
    describes how to use the script.
  - Released a source code patch for MPlayer-1.0pre6a to enable Dirac play
    back support.
  - Released DirectShow filter for Dirac to enable playback of Dirac bitstreams
    using Windows Media Player
  - Reinstituted bi-directional prediction
  - Added support for multiple quantisers.
  - Code restructure and tidying.
  - Updated bitstream version. Encoder/Decoder apps in this release will not
    be able to decode bitstreams generated by earlier versions.
  - Patches applied: 1081736,1081743

dirac_0.5.0
===========
  - Added C interface for the encoder application. Refer to the api 
    documentation in dirac_encoder.h or in the Programmer's Guide at
    http://dirac.sourceforge.net/documentation/code/programmers_guide/index.htm
  - Updated bitstream version. Encoder/Decoder apps in this release will not
    be able to decode bitstreams generated by earlier versions.
  - Improvements to Constant Quality encoding engine to improve stability.
  - Speed-ups to motion compensation and wavelet transform for faster decoding
  - Removed rounding errors from motion compensation.
  - Fixed bugs with clipping and encoding black or white areas.
  - Fixed bug with overflow in context modelling for arithmetic coding of 
    HD pictures.
  - Added unit tests for motion compensation classes and wavelet classes.
  - The number of frames parameter is no longer required. All existing picture
    header files must be recreated using make_header for this parameter to be
    omitted. Otherwise, incorrect bit rates will be reported by the encoder
    application.
  - Updated algortihm documentation.
  - Restructured libraries so that only two libraries are generated - an encoder
    library and a decoder library. 'make install' will now install only the 
    public API libraries and header files, the encoder and decoder apps and the
    conversion and instrumentation utilities.
  - Added support for building shared and/or static libraries using libtool
    under UNIX, GNU/Linux.  Shared and static libraries are built by default.
  - Added support for building shared or static libraries under MS Windows
    using MSYS and the no cost MSVC++ 2003 toolkit. Shared libraries are built
    by default.
  - Added support for building the codec using MS Visual C++ .NET 2003 under
    MS Windows. Options are available to build Dlls and static libraries. 
    The _declspec mechanism is used to export the Encoder and Decoder public 
    "C" API.
  - New target 'valgrind-check' added to make under Linux.
  - Build under MS Windows using nmake no longer supported.
  - Bug fixes: 1050816, 1055692, 1050757, 1044503, 1044501, 1044499
  - Patches Applied: 1055714, 1055707, 1061923, 1071429, 1059342

dirac-0.4.3
===========
  - Quality metric has changed from PSNR to one based on 4th powers of errors.
    This results in pictures with more stable quality and fewer artefacts.
  - Fixed bug with scaling motion vectors for non 4:4:4 formats.
  - Fixed bug in encoder in downconverting a picture with odd valued dimensions
  - Fixed bug in encoder in handling input with 4:1:1 chroma format.
  - Fixed bug in diagnostics tool in handling input with 4:1:1 chroma format.
  - Updated algorithm documentation

dirac-0.4.2
===========
  - Fixed bugs in writing instrumentation data while encoding which
    caused the diagnostics tool to go into an infinite loop
  - Fixed bug in diagnostic tool when handling reference 2 and only
    one reference frame is available
  - Number of frames in encoded output header is now set correctly to
    the number of frames encoded.
  - Fixed bug with bit-rate being miscalculated when only a section of the 
    sequence is encoded.
  - Fixed bug with setting DC band perceptual weight.

dirac-0.4.1
===========
  - Added support for build under Microsoft Windows using gnu autotools,
    MSYS and no-cost MS VC++ 2003 compiler

  - Fixed bug in building unit tests when older versions of cppunit are 
    installed

  - Fixed bugs in measurement of PSNR and motion estimation process

dirac-0.4.0
===========
  - Added constant quality encoding interface using PSNR
     as the quality measure, and using multi-pass frame
     coding
  - Added Intra-frame insertion and cut detection
  - Added C interface for the decoder application to allow
     interfacing to player software. Refer to the api documentation
     in dirac_parser.h.
  - Implemented unique frame headers to facilitate bit-stream
     parsing
  - Added instrumentation output from encoder and an 
     instrumentation display tool for visualising encoder
     decisions to help with algorithmic development
  - Cleansed and refactored numerous code files to add comments,
     improve readability and modularity
  - Added a unit testing framework based on cppunit for testing
     classes 
  - Fixed bug with vertical block separation always being set
     to 8 pixels
  - Added feature to encode only a section of input video rather
     than the whole sequence
  - Tweaked motion estimation parameters to improve performance,
     mostly for higher-res pictures
  - Included document describing the syntax of Dirac bitstream v0.1
    in the doc directory.
  - Fixed bug 1008694 (Out-of-tree building fails)
  - Fixed header and pkconfig installation
  - Updated algorithm documentation

Dirac-0.3.1
===========
  - Fixed compiling under Windows
  - Video conversion utilites now build under Windows
  - Colour conversion matrix coefficients corrected in conversion utilites

Dirac-0.3.0
===========
  - Added conversion utilities for going to and from RGB and 
     various YUV formats, plus bitmaps
  - Fixed serious artifacts when frame dimensions are not multiples of 32
  - Performance speed up writing output to disk
  - Changed .hdr format from binary to text for cross-platform interoperability
  - Added autotest framework with simple colour-bars test
  - Added format conversion, MPlayer and ImageMagick examples to README
  - Added decoder support for arbitrary temporal prediction structures
  - Added support for I-frame only coding
  - Added support for coding with a single initial I-frame
  - Extended code documentation to all major classes

Dirac-0.2.0
===========
  - Fixed crash when frame dimensions are not multiples of 32
  - Removed dependency on XParam library
  - Removed libtool for faster compiles and ease of debugging
  - Included doxygen documentation of major classes
  - Initial "programmer's guide" documentation - doc/dirac_algor.tex

Dirac-0.1.0
===========
  - Initial Sourceforge release.
