Research On Parallel Computation Of H.264 Based On CUDA

With the rapid development of High-definition technology, the heavy time cost of Multimedia coding is very long. Due to the high strengths of H.264 standard in transportation properties and storage space, it has been exploited widely in more and more video files, which provides us with a better performance. However, this trend brings about the huge increase of computing complexity in the codecs. Therefore, many research efforts have been paid to solve this problem in order to effectively reduce the computing workload of H.264 standard.As we known, There is a powerful Graphic Processor Unit(GPU) in each computer now. The GPU was initially designed to help the Graphic Render, however, its main responsibility has gradually changed from Graphic accelerating to common computing. It also supports many applicable platforms, such as the CUDA platform. In this way, as the video coding encounters complex computing, it would be a nice solution to transfer the workload of parallel computing part to the GPU with the assistant of CUDA.In this paper, the current development of H.264's key technique research was introduced at first, and the details of H.264 and CUDA architecture was proposed subsequently. Because the motion estimation and loop filter are the most complex computing modules in H.264 standard and there are substantial parallel computing workload in these two modules, these two modules are the focus of my research. The analysis of parallel computing parts is given and a parallel realization solution based on CUDA is proposed respectively for those two modules. According to the analysis of experiment results, these two solutions achieve a higher performance, although they could be improved further later.According to this paper, it proved that the performance of video encoding and decoding modules could be improved significantly with the usage of the GPU's powerful parallel computing capacity. This solution could not only ensure image quality, but also greatly improve the speed of encoding and decoding operations and could be used in many applications in future.