Neural network-based arithmetic coding of intra prediction modes in HEVC

摘要

In both H.264 and HEVC, context-adaptive binary arithmetic coding (CABAC) isadopted as the entropy coding method. CABAC relies on manually designedbinarization processes as well as handcrafted context models, which mayrestrict the compression efficiency. In this paper, we propose an arithmeticcoding strategy by training neural networks, and make preliminary studies oncoding of the intra prediction modes in HEVC. Instead of binarization, wepropose to directly estimate the probability distribution of the 35 intraprediction modes with the adoption of a multi-level arithmetic codec. Insteadof handcrafted context models, we utilize convolutional neural network (CNN) toperform the probability estimation. Simulation results show that our proposedarithmetic coding leads to as high as 9.9% bits saving compared with CABAC.