Design Of Parallel Algorithms For Several H.264Modules And Implementation Based On CUDA

With the development of Internet and information technology, video has already become a part of people’s life. Due to its large amount of information, it is hard for video to store or transmit directly. To solve the problem, people have spared no effort in the research of video compression. As a new generation of video coding standard, H.264has excellent performance, but its performance is at the cost of increasing the computational complexity.In recent years, with the rapid development of graphic processing unit, GPU has not only been applied in graphic rendering but also in general computing. Because of its low cost, flexible in programming and powerful ability in parallel data processing, GPU has been applied in the field of video coding and decoding.In this paper, we make task partitioning for the serial encoder and use a new encoder framework in which CPU and GPU work cooperatively. CPU is mainly responsible for reading of the raw data, encoding of the parameter set and slice header as well as the transmission between host and device while GPU is responsible for such modules as intra prediction, inter prediction, entropy coding and loop filter which are rich in calculation.In this paper, we mainly design parallel algorithms for intra prediction, entropy coding as well as loop filter and implement them on CUDA. We propose three parallel algorithms for intra prediction. For entropy coding which is hard to design a parallel algorithm, through analyzing the data correlation, we propose a rational parallel scheme which includes information statistics, code stream producing and combination. As for the loop filter, we present two parallel schemes and some optimization methods.Experimental results have shown that the encoder framework which CPU and GPU work cooperatively adopted in this paper can fully take advantage of the processors respectively. In the condition of ensuring the performance of compression and image restoration quality, the computational efficiency improved significantly. The overall encoding time in parallel is about5-7times faster than the popular x264serial encoder.