VQA Video Decoder.
- Mike Melanson (firstname.lastname@example.org@.email@example.com@firstname.lastname@example.org)
- See Also
The VQA video decoder outputs PAL8 or RGB555 colorspace data, depending on the type of data in the file.
This decoder needs the 42-byte VQHD header from the beginning of the VQA file passed through the extradata field. The VQHD header is laid out as:
bytes 0-3 chunk fourcc: 'VQHD' bytes 4-7 chunk size in big-endian format, should be 0x0000002A bytes 8-49 VQHD chunk data
Bytes 8-49 are what this decoder expects to see.
Briefly, VQA is a vector quantized animation format that operates in a VGA palettized colorspace. It operates on pixel vectors (blocks) of either 4x2 or 4x4 in size. Compressed VQA chunks can contain vector codebooks, palette information, and code maps for rendering vectors onto frames. Any of these components can also be compressed with a run-length encoding (RLE) algorithm commonly referred to as "format80".
VQA takes a novel approach to rate control. Each group of n frames (usually, n = 8) relies on a different vector codebook. Rather than transporting an entire codebook every 8th frame, the new codebook is broken up into 8 pieces and sent along with the compressed video chunks for each of the 8 frames preceding the 8 frames which require the codebook. A full codebook is also sent on the very first frame of a file. This is an interesting technique, although it makes random file seeking difficult despite the fact that the frames are all intracoded.
V1,2 VQA uses 12-bit codebook indexes. If the 12-bit indexes were packed into bytes and then RLE compressed, bytewise, the results would be poor. That is why the coding method divides each index into 2 parts, the top 4 bits and the bottom 8 bits, then RL encodes the 4-bit pieces together and the 8-bit pieces together. If most of the vectors are clustered into one group of 256 vectors, most of the 4-bit index pieces should be the same.
Definition in file vqavideo.c.