Research On Fast Video Transcoding Algorithm From H.264 To HEVC

The H.264 standard has been widely applied in broad variety of application domains due to the remarkable performance in flexibility and coding efficiency. However, the emergence of beyond-HD video and the limited network bandwidth resorts to a coding standard with higher compression efficiency. The new generation video standard HEVC(High Efficiency Video Coding) saves up to approximately 50% bit rate compared to H.264. The research on transcoding algorithm from H.264 to HEVC is carried out for their transition.To reduce the complexity of HEVC intra encoder, the support vector machine is applied to provide the proof for the splitting of coding unit based on the statistical analysis of residual bits, partition mode and sum of absolute difference in H.264. Meanwhile, the number of intra prediction mode is reduced by exploiting the correlation exited in H.264 and HEVC. Then, the splitting mode of inter coding unit is decided by k-means clustering, based on the statistical learning about four dimension vector which is formed by coding bits of the four sub-blocks. In addition, through the analysis of motion vector, the candidates of prediction modes are reduced and the motion process is optimized. At last, aiming to avoid the time-consuming machine learning and clustering, this thesis decides the splitting of coding unit by the partition type of blocks, whose correlation to image complexity is based on. Through the threshold, coding efficiency is traded off for the time complexity, which caters to diversity of services.Experimental results show that the proposed algorithm could save up to 43% of encoding time and maintain the similar rate-distortion performance, which could be used in the practical application.