Context-based binary arithmetic coding with stochastic bit reshuffling for advanced fine granularity scalability

We present a context-based binary arithmetic coding with a stochastic bit-reshuffling scheme to improve bit-plane coding of MPEG-4 fine granularity scalability (FGS). Current approach has not fully considered the correlations across bit-planes and among adjacent transform blocks. To fully exploit correlation, we construct context based on spatial correlation and energy distribution of transform blocks. Furthermore, we consider the context across bit-planes to reduce side information. In addition, when the enhancement-layer is partially decoded, the block update in a raster scan causes unbalanced quality of a decoded frame. To have consistent subjective quality, we reshuffle the bit transmission order with a probability model. Since both encoder and decoder have identical probability model, no extra overhead are required. For faster convergence of statistic adaptation, we replace the traditional 8×8 floating point DCT with 4×4 integer transform. As compared to MPEG-4 FGS based bit-plane coding, we can improve the PSNR by 0.5∼1 dB and offer better subjective quality. Specifically, our approach is among top 3 of the test in the subjective quality evaluation of scalable video coding in the 65th MPEG meeting in July 2003.