AIC - Future Work
The Advanced Image Coding codec is still experimental and in no way a finished
product. My initial goal was not to make the codec too complex. The following
improvements could make the AIC codec perform even better, at the cost of making
it more complex:
- The H.264 standards includes a built-in deblocking filter to smooth the
edges of the 4x4 image blocks. Incorporating this filter (or a modified version)
into AIC may improve the visual quality of the image, especially at lower
bit rates (although it might lower the objective image quality).
- The AIC codec can currently not be used for lossless image coding because
of the rounding errors in the color conversion and Discrete Cosine Transform
algorithms. To make AIC suitable for lossless coding, reversible color conversion
and transformation algorithms must be incorporated. The JPEG-2000 standard
includes a reversible color transform that might be suitable for AIC. There
are also some reversible DCT transformations around (for example the BinDCT
algorithm) that can be used for lossless image coding. Or maybe an integer
wavelet transform is more appropriate.
- The H.264 standard defines 9 prediction modes that are reasonably fast to
compute. When encoding speed is not a top priority, maybe some other prediction
modes that interpolate between the left and top edge can be used. These are
more computationally intensive, but may produce better results. On the other
hand, increasing the number of prediction modes also increases the number
of bits needed to encode the prediction mode.
These improvements are under current investigation. You are invited to make
your own improvements by downloading the source code.
Just let me known about your findings.