Research On Video Compression Algorithm Based On Multi-core Computing Platform

Based on research about H.264/AVC video compression standard and multi-core computing platforms, this paper deeply studied the parallel granularity and execution processes of the core algorithms of H.264/AVC, such as intra-prediction, motion estimation, DCT transformation and entropy coding. We mapped these traditional sequential core algorithms to NVIDIA CUDA parallel computing model, and assessed the speedup of the GPU's implementations.Experimental results showed that multi-core computing platforms can significantly speed up the H.264/AVC encoding. For example, FS motion estimation, which has very high calculation intensity, archives the best performance and the speedup is up to 160. It's known that GPU is more suitable for computing-intensive process, and control flow statements will greatly affect the execution performance, but a large number of parallel threads can compensate the impact of branch operations. For example, the speedup of intra-prediction, fast motion estimation and CAVLC entropy coding are 9, 3 and 3, respectively.