Research And Implementation Of HEVC Parallel Encoding Algorithm Based On Tilera Multi-core Processor

HEVC is the new international standard for high and ultra-high definition video coding,providing higher compression performance. However, the extreme high computational complexity becomes a major obstacle in application. With the development of multi-core processing architecture, parallel encoding in multi-core platform presents an effective way. Based on exploring the potential parallel processing of HEVC, as well as the multi-core mechanisms, the thesis conducts study on parallel HEVC video coding techniques on TILE-Gx36 multi-core platform, and proposes three parallel coding algorithms. The main work and innovations are as follows:(1)Inspired by WPP method in HEVC, we investigate the intra- and inter-CTB prediction dependencies to its reference CTBs, and propose a parallel HEVC encoding scheme based on intraand inter-CTB row. Using the inter-correlation, we can encode the CTB row of the next frame when the CTB rows in current frame need not to process. This parallel method solves the problem of WPP and improves parallel speedup significantly.(2)Due to the high computational complexity of the whole traversal algorithm used by HEVC intra-prediction, which affects the intra-prediction encoding speed. In order to improve the coding efficiency, we research the dependencies of CUs at different depths, and propose a new parallel intra-prediction algorithm. Using priority queue, we can encode CUs at different depths in parallel, while in 4×4 pixel size, by sharing the reference pixels, two 4×4 CUs can be processed at the same time, which makes a great improvement in encoding speed.(3)Based on the above research, a multi-granularity fusion parallel HEVC encoder scheme,which combines CTU level parallelism and CU level parallelism, is proposed. Furthermore, a multi-core distribution and multi-thread scheduling strategy is designed to ensure the full use of multi-core resource, meantime, semaphore and mutex are introduced to synchronize and communicate the processing units. At CTU level, we use the intra- and inter-CTB row parallel algorithm to achieve intra- and inter-CTB rows encoded in parallel. And at CU level, we use parallel intra-prediction method coding different CUs simultaneously.Experiments are based on high definition and ultra-high definition test sequences, the speedup of CTB row level parallel algorithm reaches to 6.2, the parallel intra-prediction algorithm saves74% prediction time, and multi-granularity fusion parallel HEVC encoder scheme break through the bottlenecks of the CTB row level parallel algorithm that its speedup reaches to 6.3. Experimental result shows that, under the premise of not degrading the video quality, our proposed method reduces the computational complexity, speed up the encoding speed and improves parallel speedup significantly.Finally, the thesis gives a summary of the work and prospects the future research.