Research On GPU-based Parallel Intra-frame Coding Mechanism Of HEVC

With the growing popularity of digital video, video encoding technology is particularly important. As one of the most main links of video coding technique, the importance of intra coding can be imagined. GPUdevice has developed rapidly in recent years and its performance has big enhancement, but intra prediction in HEVC(High Efficiency Video Coding) cannot take good use of the powerful computing performance of GPU because algorithms parallelism are limited. Most of The state-of-the-art researches in HEVC only focus on improving the algorithms of some function modules, or implementing coarse-grained parallelism such as block-level and chip-level parallel. But little research involves the further fine-grained parallelism, such as pixel level.Fine-grained parallel intra coding mechanism based on CUDA architecture realizes the pixel level and parallel in prediction blocks on the basis of block level parallel. It increases the parallelism and greatly improves the coding efficiency with the help of fast block and mode decision. The mechanism uses a variety of optimization strategies: Firstly, unify and simplify intra prediction formulae, establish a unified reference array, and propose the look-up table method to implement fine-grained parallelism. Secondly, only two block sizes, 8×8 and 32×32, are chosen during intra prediction process to simplify the calculation under the premise of little effect on the coding performance. Thirdly, optimize mode decision process to reduce the number of candidate modes, which need calculate the rate-distortion cost and reduce the encoding time. Finally, improve the coding sequence of to improve coarse-grained level parallelism, and then combine with fine-grained parallel to further improve the parallelism as well as enhance coding efficiency.Using video test sequence with the resolution of 3840*2160, 2560*1600 and 1920*1080 in the above intra coding mechanism system, the average coding efficiency is from 10 times to17 times than the standard serial intra coding mechanism. The proposed fine-grained parallel intra coding mechanism based on CUDA is stable and effective.