Research On Regional Context Model And Dynamic Binarization In Entropy Coding

The application of super high-density video has been widely spread inthe field of mobile communication and multi-media on network. To meetthe demand, HEVC made the official public release which imported moredirection contexts, truncated rice in the binarization procedure of CABACand other algorithms based on H.264/AVC.To further improve the video compression efficiency by appropriatecontext model and dynamic adaptivity, both the regional context modeland dynamic binarization in entropy coding, which are in the procedure ofcontext modeling and encoding, are researched by this paper.There are a lot of context model algorithms which make theprobability statistics by the context model, such as context tree weightingand prediction with partial matching. When applied to multi-directioncontext model, these context model algorithms make the modelredundancy double-exponential increase.To solve the problem in multi-direction context model, this paperadopts the algorithm which centers the current pixel to be predicted, sortsthe pixels in the region according to the line distance from the center pixeland calculates the normal coefficient by the multiple of the amount that ismax line distance minus current line distance, to the power of the count ofthe pixels which have the same line distance. This algorithm takes accountof the impact of prediction accuracy from both the line distance and thecount of pixels with the same line distance.In the experiments of the compression for the test sequences ofH.264/AVC, it turned out that the region context model improved thecompression efficiency. To reduce the count of multiple operation in arithmetic coding,CABAC made the state machine which stood for some probability. In theprocedure of encoding, it will change the state from0to63. Theprobability of next state is the multiple of the current probability and scalefactor. In the encoding process, the real probability of the symbol may beout of the range for the state probability. In this case, the encodingredundancy will be produced and the multi-bins for binarization willenlarge the redundancy.This paper applies dynamic Huffman encoding to update binarization.At each turn of encoding a symbol, this algorithm updates the frequency ofthe node from the leaf node to the root and checks the balance of eachnode.In the experiments of the Java file compression, dynamic binarizationimproved the compression ratio for the data with high probability. By theexperiments of the DNA sequence data compression, dynamic binarizationwill also improved the compression ratio.