Research Of Multi-view Video Coding Based On Fast Walsh Transform

With the development of information science, traditional ordinary 2D video has already can’t satisfy people’s visual requirements. The appearance of 3DTV, HDTV, FVT meet the needs of people to get a lively scene, they can bring the immersive visual perception to the viewers. The trend of multimedia communication technology has been developed from ordinary 2D video to multi-view video. The multi-view video is captured by a series of cameras in different positions at the same time, because of the different shooting angles, people will have a stereoscopic perception when watch it. Now 3D movie is the most widespread multi-view video, it is two-view video, which uses the different perspective of human eyes and aggregation function.It will take hundred Mbits to store ordinary 2D video, sometimes up to Gbits, and multi-view video contains several single-view videos, the data of multi-view video will increase exponentially, and it brings major difficulties for the storage and transmission of multi-view video. Considering the redundancy and the correlation of video sequence, and multi-view video itself contains a large amount of redundant information, so the multi-view video can be compressed.The research of ordinary 2D video coding is relatively mature, but the study of multi-view video coding is at primary stage. At present, there are already many algorithms about multi-view video coding, but most of them are based on MPEG-4 or H.264 combined with motion compensated prediction(MCP) and disparity compensated prediction(DCP). One channel is the main view, and coding with MPEG-4 or H.264, the other channel is the auxiliary channel, and coding with MCP and DCP.Combined with the characteristics of video sequence, our lab put forward the multi-dimensional vector matrix theory. The multi-dimensional vector matrix theory model is extended from traditional 2 dimensions to M dimensions. Due to the multi-dimensional vector matrix, the multi-dimensional data of multi-view video can be expressed in one model, so it can effectively eliminate the redundancy. In this paper we proposed the multi-dimensional Walsh transform nuclear matrix which based on the multi-dimensional vector matrix, and verified its orthogonality. The Walsh transformation can effectively remove the correlation of multi-views, and further compress the data. Therefore, the purpose of this paper is to achieve the multi-view video coding which based on fast Walsh transform, and obtain some experimental results on condition that ensuring the reconstructed video quality.In the experiment, we first block the original multi-view video, considering the correlation of every views, we reorganize the blocking data by frame order. Combined with the multi-dimensional vector matrix theory, Walsh transform is done after reorganization. Quantization is used in this paper to further compress the data, we abandon some information which has little impact on subjective video quality. After quantization, scanning is performed, finally run length coding(RLC) is used to encode the quantized data. The decoding process is the inverse of the encoding process.The experiment is done in 8×8 blocking, and use CR, PSNR and the whole time of encoding as the performance evaluation criteria. Compared with the original multi-view video, the reconstructed multi-view video quality is good, and proves the feasibility and the effectiveness of this method. The advantage of this paper is that we use the characteristics of the Walsh transform nuclear operator which is separable and fast, and the 4-dimension transformation can be reduced to 1-dimension, the computation complexity is cut down and the computation efficiency is increased. Our algorithm is performed in the simulation environment, if combined with the embedded platform, there will be better results.