Context-Adaptive Binary Arithmetic Coding With Fixed-Length Codewords

Adaptive binary arithmetic coding is a widespread technique in the field of image and video coding.Most traditional arithmetic coder only generates a codeword with a priori unknown length.The generation of this codeword requires a renormalization procedure to process the symbol sequence progressively.Therefore,the conventional arithmetic code has high computational complexity and the code is not partially decodable,there is almost no robustness.This paper introduces a FLW arithmetic coder,Different from the traditional method,FLW avoids the procedure of renormalization and reduces the computational complexity,the encoder can generate a plurality of fixed length codewords independently,so it can also be regarded as a block arithmetic coding,and each codeword can be independently decoded as needed.After the codeword is interfered by noise during transmission,he error generated by the decoding does not affect the decoding result of the next codeword,which greatly improves the robustness.However,while the FLW codec solves the shortcomings of the traditional arithmetic codec,the compression effect is worse than that of the conventional arithmetic coding.Therefore,this article adds a Context model to the FLW codec to improve compression efficiency.Considering the correlation between coded symbols in the source sequence,the conditional probability of the symbol sequence is used instead of the probability of unconditional symbol probability,to perform a context modeling.In theory,the higher order of the context model,the better the compression will be.However,when the order of the context model is too high,the estimated conditional probability distribution at this time is too many and the number of samples in each distribution is less,which does not perform well in statistical characteristics,so it can't get the expected compression gain.Therefore,in this paper,the statistical conditional probability distribution is used to calculate the adaptive codelength increment after the conditional combination in high-order context model,to quantify the model.Producing a new conditional probability distribution with better statistical properties,which fully exploit the statistical properties of source symbol sequences.Experiments show that,applying the quantized Context model to the FLW arithmetic coding,the impact on the compression effect is very small,while improving these shortcomings of traditional arithmetic coding.