Arithmetic coding is employed in image and video coding schemes to reduce the statistical redundancy of symbols emitted by coding engines. Most arithmetic coders proposed in the literature generate variable-length codes, i.e., they produce one long codeword of variable size. This requires renormalization operations to control the internal registers of the coder and the propagation of carry bits. This paper introduces an arithmetic coder that generates fixed-length codewords. The main advantage of the proposed coder is that it avoids renormalization procedures, which reduces computational complexity. Also, it uses a variable-size sliding window mechanism to estimate with high precision the probability of the emitted symbols. Experimental results indicate that the proposed coder achieves coding efficiency superior to those coders employed in JPEG2000 and HEVC while having lower computational costs. When integrated in a JPEG2000 implementation, the proposed coder achieves coding gains between 0.5 to 1 dB at medium and high rates, and speedups between 1.1 to 1.3 in the bitplane coding stage.
|Original language||American English|
|Number of pages||5|
|Journal||2014 IEEE International Conference on Image Processing, ICIP 2014|
|Publication status||Published - 28 Jan 2014|
- Arithmetic coding
- context-adaptive models