On Distributed Video Multi-Hypothesis Reconstruction Technology Based On Compressed Sensing

With the rapid progress of modern technology,people's demand for video media is increasing,which brings about a huge amount of data storage and transmission.The theory of compressive sensing can overcome the limitation of the Nyquist frequency,allowing to sample at less than 2 times the Nyquist frequency and more and more widely concerned by the academic community.Distributed video compressed sensing technology combines compressed sensing technology with distributed video coding technology,which reduces the bandwidth requirement of transmission signals.The traditional distributed video compressive sensing framework shifts the complexity from the encoding end to the decoding end,and adopts the framework of "independent coding and joint decoding" to reduce the complexity of the coding end.The article analyzes the research status of sparse adaptive matching pursuit algorithm and distributed video compressive sensing at home and abroad,and study related theories such as compressive sensing reconstruction algorithm and distributed video compressive sensing multiple hypothesis reconstruction algorithm.In view of the step size design of existing compression sensing algorithms is not reasonable,low accuracy,as well as the existing distributed video compression sensing to assume reconstruction solutions has failed to make full use of non-local similarity data of the video frames,The article carries out research on distributed video multi-hypothesis reconstruction technology based on compressed sensing,focusing on variable-step-based compressed sensing reconstruction algorithms,Multi-hypothesis reconstruction project of DCVS based on weighted Non-local Similarity,and Distributed Compressive Video Sensing Multi-Hypothesis reconstruction algorithm based on Corner Detection.The main innovations and work in this article are as follows:(1)A sparsity adaptive matching pursuit algorithm based on optimal step size is proposed.In order to solve the problem of unreasonable step length in SAMP(Sparsity Adaptive Matching Pursuit).In order to solve the problem of unreasonable step length in SAMP(Sparsity Adaptive Matching Pursuit),an optimized step size is designed and introduced into the algorithm.Experiments show that compared with the original algorithm,this algorithm has a lower reconstructed MSE and higher precision reconstruction probability.Compared with the original SAMP algorithm,it also improves the reconstruction time by 20%.(2)SAMP algorithm based on Parabolic Variable step for Regularized Backtracking(SAMP-PVRB)is proposed based on variable step size.For the problems of overestimation and underestimation in the SAMP-RB algorithm,parabolic steps are introduced into it.Experiments show that the proposed algorithm further improves the Mean Square Error and the accurate reconstruction probability.Compared with the original algorithm,the reconstruction time is shorter than that of the original SAMP-RB by 20%.(3)A Multi-hypothesis reconstruction algorithm of DCVS based on weighted Non-local Similarity is proposed.Aiming at the shortage of original video compressed sensing scheme,the weighted nonlocal similarity is introduced into the existing scheme.The simulation experiments on standard video sequences show that the improved algorithm PSNR(Peak Signal to Noise Ratio)increases the average 0.2-2d B,has a higher SSIM(Structural Similarity Index)index and a better visual reconstruction effect.(4)Distributed Compressive Video Sensing Multi-Hypothesis reconstruction algorithm based on Corner Detection is proposed,aiming at the problem of low similarity for non-locally similar blocks,and further propose an improved algorithm.Use corner detection to further filter out high-precision similarity blocks.Simulation experiments on different video sequences show that the improved algorithm has higher PSNR index and better visual reconstruction effect,and effectively solves the problem of distributed video compression with low non-local similarity block similarity.