The Research Of Video Transcoding Algorithm

With the development of the multimedia communication and the video coding technology, several video coding standards have been established for various video applications. As the number of networks and the coexistence of the video standards, interoperability between different systems and different networks is becoming more important. Video transcoding is an efficient way to solve this problem. A video transcoder is a device that can convert video from one compressed format into another compressed format with different bitrate, resolution or syntax. This thesis focuses on the research of video transcoding algorithm.Firstly, the key problem of inter-standard transcoding technology is presented. By analyzing the cause of drift error in the texture and motion vector reusing transcoding architecture from MPEG-4 to H.264, an error compensation method is proposed. In the rate control part, the basic unit level bit allocation algorithm is emphatically studied. The allocated bit number for the re-encoded basic unit is allocated in term of the information of the pre-coded bitstream and the allocated bit number actually reaches approximately the target bit number.In the spatial resolution transcoding, in order to reduce the computation complexity, we present an fast integer transform-domain intra frame downscale transcoding algorithm. A method to estimate the edge direction inside block in the integer-transform domain is derived. Based on the statistical analysis of the relationship of the variance of the sixteen 4×4 sub-block DC coefficients inside one macroblock and the intra prediction size, a prediction size selection algorithm is proposed and a concept named "transition interval" is introduced. The wrong judgement number can efficiently be reduced through setting "double threshold" for the prediction selection algorithm. There exists an relativity among the intra prediction modes of the neighboring blocks, we give a judgement rule to predict the current block coding mode from the neighboring blocks, efficiently reducing the candidates of the prediction mode. In inter frame downscale transcoding, two methods to predict the coding mode of current macroblock are proposed: mode-mapping method and residual-prediction method. Then, we use the motion vector diversity of the corresponding mapping area to further reduce the candidates of the inter prediction modes. The experiment results show that our method can reduce the number of candidates of the coding modes obviously and maintain good picture quality meanwhie.In the last, we study the existing motion vector composition methods and propose an improved ADVS motion vector composition algorithm considering the characters of the MPEG-4 to H.264/AVC transcoding. Then, a method to predict the coding mode of the current macroblock is proposed using its coding mode in the coded bitstream and its reference macroblock coding mode in the dropped frame, thus the range of the inter-mode searching is reduced by 50 percent. The experiment results show that the PSNR and bitrate of the algorithm are close to those of the re-encoding results.