Context-based binary arithmetic coding for fine granuality scalability

Abstract

We proposed a context-based binary arithmetic coding to improve entropy coding of the enhancement layer for the H.26L-, based Fine Granularity Scalability (FGS). Instead of independent bit-plane coding, we exploit correlation between various bit-planes and neighboring coefficients. For each bit-plane, we perform significant/refinement bit partition and construct contexts separately in terms of energy clustering and distribution of DCT coefficients. For the significant bit, we extend one-dimensional run-length coding to multiple dimensions and segment the significant bit-planes. For the refinement bit, we put together refinement bits exhibiting stronger correlation by energy groups and apply run-length coding distinctively. Particularly, we introduce the maximum run concept to offer better statistical adaptation. Averagely, our approach reduces the enhancement-layer bit stream by 7similar to8% and improves the PSNR by 0.3similar to0.5dB. Moreover, our performance outperforms the JPEG-2000 for the encoding of the enhancement layer.