Context Probability Modeling for Encoding Quantized Transform Coefficients

Probability modeling of the source statistics is a key step that decides the compression efficiency of entropy coders. Both CAVLC and CABAC, the two entropy coders in H.264/AVC, use Markov models. For coding moving pictures, quantized transform coefficients take up more than 50% of the total bit rate. In this paper, we compare the probability modeling process of quantized transform coefficients in CAVLC and CABAC, and analyze their compression efficiency. We also explore in depth the statistic character that can be used to improve the coding efficiency. Then we propose our new classification method to build Markov models. Experimental results show that averagely 3% bit rate of quantized transform coefficients can be saved, even more bit rate savings for large picture resolution and Intra frames. And our method doesn’t obviously increase the complexity for either software or hardware implementation. Index Terms — entropy coding, probability model, Markov model, CAVLC, CABAC, quantized transform coefficients, H.264, MPEG-4 AVC