The Parallel Design Of H.264 To AVS Video Transcoding Based On GPU

H.264 is the most optimum video coding standard so far, but the high patent fees hinders the widespread use in China, therefore we have self-developed AVS standard, which almost has the comparable performance with H.264. The two standards will coexist for a long time in the future, however the high computational complexity has become an obstacle to the development of video transcoding.In recent years, GPU (Graphic Processing Unit) increases rapidly in computing power, especially NVIDIA has introduced the CUDA software platform supporting the C language and OpenCL in 2007, which making the GPU application development simple, the GPU used in the field of video transcoding will have very broad application prospects.The parallel transcoder uses the two dual-thread of CPU and GPU, which are working currently. CPU is responsible for the entropy decoding, data exchange and resource scheduling in the transcoding as main thread, GPU is responsible for the other works of transcoding expect the entropy decoding and modules of work as sub thread, this design can reduce the number of data transfer between CPU and GPU. In order to make full use of the parallel computing of GPU, the parallel algorithms for the compute-intensive and time-consuming modular are proposed in this paper. Because of the correlation in adjacent blocks in the intra prediction, the ladder parallel algorithm is designed in this paper. For the inter-frame motion estimation, a parallel algorithms for motion estimation based on full search is proposed, this algorithm set the starting point to 0, canceling the step of using the adjacent block motion vector to predict the starting point. Because of many challenges in entroy coding, such as data dependencies, control-related and stream stored priority, a parallel algorithm of entroy coding is proposed in this paper, which splits entropy coding and storage priority stream correlation between the macro-blocks and solving the problem limiting the parallelism. Because of the similarity between two video standard, the parallel intra prediction and parallel motion estimation algorithms of reusing the information from the decoder were proposed in this paper, which can reduce the computational complexity to increase the speed of the transcoding.Finally, many experiments are done to test image perforemance, acceleration of each module in transcoding process and the performance evaluation of overall transcoder. The experimental results show that the parallel trancoder proposed in this paper can make great advantage of the CPU and GPU hardware features, providing reference for parallel applications in other areas.