1. Color in BTPC is processed as RGB for lossless and YUV (4:1:1) for lossy. Currently there is no between-color-plane information used in the predictors. It is possible to do much better than this for graphics, where the dependency between planes is very strong. This is particularly the case for lossless coding. Furthermore, expansion of the UV planes in the decoder is performed without reference to the Y plane. Again, there is the possibility of exploiting between-plane information to generate a better interpolation. Note that color-coding cannot use a palette because performance on multimedia images must be maintained. The evaluation code is set up so that later versions with enhanced color performance will continue to be called BTPC2, and the existing decoder will be able to recover at least the Y component.
2. The decoder (which is the "optimized" half of BTPC 2) is somewhat slower than JPEG decoding with the Independent JPEG group's superb version 5. The bottleneck is the adaptive huffman decoding which involves walking a tree for every input symbol. BTPC3 will use some other approach to Huffman coding (almost certainly with a slight loss in compression) to ensure that decoding is very fast. Possibly static code tables for each band will be calculated in the encoder and prepended to the data.
3. BTPC is being applied to moving pictures so that video, animations, and video with overlaid graphics can be handled efficiently.
4. Joint quantization of siblings, or of siblings and their immediate offspring (i.e. two generations' descendents from a single point) should allow increased compression at minimal cost in efficiency. The current approach is to use modified lattice vector quantizers. This might help deal with one of the problems that BTPC has at low data rates - the dropping of small but important visual features.
Next (References) | Previous (Results and discussion) | Top (BTPC)