Research And Achievement On The HD Video Processing Algorithm In GPU Environment

With the development of network and multimedia, the technology of multimedia is widely used in all the fields of society. Due to the hug information of high definition video and image, the new requirement is put froward for current processing of video and image, storing, and network transmission. There are a lot of new computing architecture and methods of high performance computing which can impove processing efficient and decrease delay. In today's computer architecture, multi-cores architecture is one of direction of development. The architecture of CPU and GPU is another development direction. Under later architecture, the research have been done how impove the parrallel for transcoding and encoding of high definition video based on theory, the new parallel propose were put forward, detail experiment and comparasion result were supplied.The main research and creative work are the following:Firstly, the new parallel proposes were put forward for HARR and 5/3 wavelet. Based on HARR wavelet theory, the parallel computing is handled by unit of four elements. Meanwhile, by unit of row, the parallel computing is put forward for 5/3 wavelet. The time of dealing with high definition vedio based on 5/3 wavelet in the proposed algorithm can decrease 90% compared with multi-cores CPU.Secondly, one parallel fast MPEG-2/H.264 intra transcoding algorithm PFIT were proposed, which is composed of three parts: by doing research on DCT coefficence, the smoothly of macroblock are analysized, and mode selection for H.264 were decided; parallel IDCT computing; by using GPU to achieve SOBEL edge detection algorithm and extract all the possible directions in 4x4 subblocks of macroblock, then make a direction statistics for 8x8 subblocks and 16x16 macroblock from the result of 4x4 subblocks. In order to improve parallel of CPU, OpenMP technology is used to control multiple GPU computation. The results of experiments show that this algorithm reduces the complexity of H.264 intra transcoding at the cost of little PSNR. Meanwhile, the time of process increase about 50%.Thirdly, based on GPU, one parallel fast MPEG-2/H.264 inter transcoding algorithm PFPT is proposesed which is composed three parts. First, by using of the coding of MPEG-2 to decide the activity of macroblock, the proposed algorithm can decide H.264's code type (SKIP, Intra, Inter), Intra mode decision and Inter splitting mode. The previous motion vector (PMV) of H.264 is decided by 1/2 motion vector of MPEG-2, with the mv size, and the H.264 adjusts searching window. Second, GPU parallel process technology decrease 4x4 SAD computation complexity of motion search process. Third, OpenMP technology is used to control multiple GPU computation. The proposed can improve 90% than standard cascade transcodingFourthly, based on the algorithm in section 2 and section 3, one GPU parallel acceleration MPEG-2/H.264 resolution transcoding algorithm PFRT were proposed. First, intra resolution transcoding depend on adjacent four macroblock, and the prediction mode is still according to SOBEL edge detection to predict corresponding modes. Second, inter resolution transcoding make decision of mode and mv by four adjacent macroblock's activity and mv. Compared with standard cascade transcoding, the computation time of the resolution transcoding improves 50% in intra, and 90% in inter.Fifthly, the transmission architecture of high definition with dual-GPUs was proposed, and the efficient on this architecture is analyzed.