Joint optimization of run-length coding, context-based arithmetic coding and quantization step sizes

Given the JPEG syntax, the rate-distortion performance a JPEG optimization method can improve is limited. Part of the limitation comes from the poor context modeling used by a JPEG coder, which fails to take full advantage of the pixel correlation existing in both space and frequency domains. Consequently, context-based arithmetic coding is proposed in the literature to replace the Huffman coding used in JPEG for better rate-distortion performance. In this paper, we extend our previous JPEG compatible joint optimization algorithm to a context-based arithmetic coding scenario. Experimental results show that an extra of 10~15% size reduction or 0.5 dB compression gain can be achieved on top of JPEG compatible joint optimization with the same level of complexity.