Creative YUV (CYUV) stream format for AVI
                 -----------------------------------------

                        Dr. Tim Ferguson, 2001.

Originally designed for the Creative Labs Video Blaster, the Creative
CYUV video codec stores frames in a DPCM coded YUV411 format.  Encoding
and decoding of frames was originally performed by the vblaster.drv
Windows driver.  This codec looks to have been designed for simplicity and
speed.

The CYUV codec is a very simple predictive coder.  A video frame is coded
in the YUV colour space, with the chrominance components subsampled by
four in the horizontal direction (YUV411).  That is, for four horizontal
Y pixels there are only one U and V pixels:
   +----+----+----+----+  +---+  +---+
   | y0 | y1 | y2 | y3 |  | u |  | v |
   +----+----+----+----+  +---+  +---+
Note this is different to the YUV420 formats used by the ITU standards
(MPEG, etc).

Predictive coding is reset at the start of each scan line, with the very
first Y, U and V component pixels being coded using the upper four bits
of their absolute values.  These provide the initial predictions for the
remainder of the scanline.  Each subsequent Y, U and V pixels in the
scanline are coded using the difference between the last decoded pixel
and the current pixel of the same component (DPCM).  This difference
is known as a prediction error (ie: the previous pixel is used as a
prediction for the current pixel).

The prediction error values are coded using three 16 entry tables of
signed values found at the start of the frame.  The first of the three
tables appears to be unused (see Note 1).  The second table is used to
code the prediction errors for the Y component and the third table is used
to code the prediction errors for both the U and V components.  These
tables allow for either linear or non-linear quantisation of the prediction
errors.

Since there are 16 entries per table, this requires four bits to index.
Thus four bits per component are used, which are packed into bytes as
follows:

  +----+----+   +----+----+   +----+----+    +--+--+  +--+--+  +--+--+
  | U1 | Y1 |   | V1 | Y2 |   | Y4 | Y3 |    |U2|Y5|  |V2|Y6|  |Y8|Y7| UY VY ...
  +----+----+   +----+----+   +----+----+    +--+--+  +--+--+  +--+--+ 
  7    4    0   7    4    0   7    4    0

Therefore, four YUV411 pixels are coded into three bytes providing six bits
per pixel.  If the above diagram were the start of a scanline, the Y1, U1
and V1 values would be coded using the upper four bits of their absolute
values (initial predictions).  The Y2, Y3, Y4,... values would then be
coded as prediction errors using table two, and the U2, U3, ... and
V2, V3, ... values would be coded as prediction errors using table three.

The overall format of a CYUV coded frame is as follows:

     +-----------------------+
     | Prediction Table 1    |  16 bytes
     +-----------------------+
     | Prediction Table 2    |  16 bytes
     +-----------------------+
     | Prediction Table 3    |  16 bytes
     +-----------------------+
     |    .      .      .    |  (width * height * 6)/8 bytes
       DPCM coded YUV pixels      
     |    .      .      .    |
     +-----------------------+

Note 1: This format was tested using the xanim (http://xanim.va.pubnix.com/)
   video player.  Therefore, the format may not be strictly compatible with
   the old vblaster.drv.  This has not been tested.

---------------------------------------------------------------------
This document was written by Dr. Tim Ferguson, 2001.

For more details, on this and other codecs, including source code, visit:
    http://www.csse.monash.edu.au/~timf/videocodec.html 

To contact me, email: timf@csse.monash.edu.au
---------------------------------------------------------------------