The Research On Distributed Video Coding

With the developing of video applications, the digital coding compression techniques get very large attentions, and come out many research outputs, which promote the whole coding performance continually. Recently, the new distributed video applications appear, such as video sensor surveillance networks, most of which need many times encoding and one time decoding. At the same time, the computing complexity and the energy is constrained. This demands that the system must have low complexity encoder and relative highly compression performance. But the traditional video coding method can’t fit the constraint, so the distributed video coding (DVC) gets experts’attention.Since in DVC, the decoder takes the function of motion estimation and compensation, which leads to the complexity of encoder lower than the decoder’s. At the same time, as DVC uses source-channel joint coding, it has very good error resilience performance, so can be used in wireless video application scenes. So DVC becomes the hot spot and the front of the international research presently.The mainly work and innovation of this paper is as below:1. Focusing on the problem that feedback channel needs extra hardware and introduces decoding delay, this paper proposed one rate estimate algorithm at the encoder side without feedback channel.Although the source rate estimation based on feedback channel can almost reach the functionally best rate for transmission, but the existence of feedback channel and the decoding delay becomes the largest hurdle for reality and real-lime applications.The concept of iteration convergence rate is introduced in the rate estimation method. Based on the research of LDPC and LDPCA, when the number of check bits comes to a threshold, in the statistic sense, the iteration will converge. The threshold is related to the degree distribution structure of the Bipartite Graphs (Tanner map) produced by LDPCA. This threshold is the lower bound of successfully iteration decoding.So it can apply the convergence rate into the source rate estimation. At the same time, it needs to consider the problem of estimation deviation for the crossover probability. If the estimated crossover probability is below the real value, this deviation may cause that the predicted source rate to be less than the real needed one, so the decoding will fail. So in rate estimate, we should save some margin for compensating the estimation deviation of crossover probability.The simulate result shows that the performance can be improved by0.35to1.7dB compared to the original algorithm.2. Focusing on the problem that the quality of side information is lower in the complex motion scenes, this paper proposed an algorithm based on block classification and OBMC and OBME liked side information generation means.To adapt to the phenomenon that there are some static or low motion regions in each video frame, this algorithm category the blocks into many modes, including skip block with almost static, direct block with simple linear motion and wz (Wyner-Ziv) block with complex motion. Through adding the direct block mode, compared to the only skip block mode, it can further saving the coding rate, and no obvious quality reducing of the video in the complex motion situation.In the side information generation, based on the principle of checking the amplitude of the estimated motion vector and the estimated residual, this algorithm in turn using the unidirectional parallel motion estimation, bi-directional symmetric motion estimation and compensation, bi-directional parallel motion compensation and the overlapped block motion estimation and compensation liked modules. The simulation shows that the performance can be improved by0.21to1.4dB compared to DISCOVER.This algorithm also researches the situation of objects moving into or out of the video frame boundary. Based on the direction of the motion vector and the amplitude of its residual, this algorithm uses the selective unidirectional parallel motion compensation means, and then combines this result with the result of previous step using weighting average method.3. Focusing on the problem that there is spatial non-stationary in the motion videos, this paper proposed one side information improving algorithm for one framework based on block classification.This original algorithm is based on the combination of distributed coding in DISCOVER, with the block classification idea in PRISM. It breaks the limits of key frame and wz frame, and takes distributed coding also for the key frame. It can adapt to the spatial non-stationary correlation relationship in the video stream. At the same time, it can take the advantages of different coding methods, and combine them to improve the coding performance.The original algorithm is using the weighted averaging algorithm based on the temporal distance for pixel to produce the side information. This algorithm can well adapt to the slowly illumination changing in the pixel and simple linear constant moving.But to regions of intense motion (although can be mostly classified into intra block type in previous operation) this averaging can’t get very good performance. Lying on this reason, this chapter advances an improved side information generation module. Based on the characters of the original framework, the new algorithm combines the block parallel motion extrapolation estimation method with the original time weighting one, and averages the two results to get the final result.For the original algorithm, as it uses the side information generating method based on the previous and subsequent intra block. The needed store capacity is relative big. So this paper researches the method of reducing the needing capacity. With a little performance loss, the storing frame number can be reduced to one or three frames.Through the simulation, the performance can be improved by0.54to1dB using the new algorithm.