Research On Block Classification Based Inter-Frame Group Sparse Reconstruction Algorithm For Compressed Video Sensing

Traditional video encoding and decoding technology based on Nyquist sampling theory,stipulates that the sampling frequency must be more than twice the highest frequency of the original signal,and the redundant information obtained by high-speed sampling should be compressed before transmitted to the decoder.This scheme requires high computaitional capacity of sampling equipment,so it is not suitable for the limited sampling environment.The Compressive Sensing(CS)technology,which was put forward in 2006,combines sampling and compressing process effectively and the compression is completed while sampling.CS theory has simplified the encoding process,and reduced the calculation burden at the encoder effectively.The theory has a great application prospect in some scenarios where the information acquisition is slow or the resources of encoder are limited.The structural similarity based inter-frame group sparsity representation reconstruction algorithm(SSIM-Inter F-GSR)is one of the best iterative optimization reconstruction algorithms for compressed video sensing.In this paper,based on the deep research of SSIM-Inter F-GSR,we propose the research on the block classification based inter-frame group sparse reconstruction algorithm for compressed video sensing.Image blocks are first classified into different types according to their characteristics,and then different improved algorithm are proposed for each type blocks respectively.The main work is divided into the following two parts.1.The block classification based adaptive threshold adjustment group sparse reconstruction algorithm(BC-ATA-GSR)is put forward.First,the image blocks are classified according to the motion state,and the reference frames for different type of blocks are selected reasonably.Then,the initial threshold for sparsing is set adaptively according to the sampling rate and the type of blocks.And during the iteration process,the threshold descends gradiently.Compared with SSIM-Inter FGSR algorithm,the proposed BC-ATA-GSR effectively improves the reconstructed videos quality and reduces the algorithm complexity.It also has significantly improved reconstruction performance compared with the latest algorithm PBCR-DCVS and the best multi-hypothesis algorithm 2s MHR.2.In BC-ATA-GSR,the sparsing process is not suitable for moving texture blocks at low sampling rates,resulting in lossing a lot of detailed information.Besides that,the similar image block matching scheme based on MSE and SSIM has limited performance.In order to get rid of these disadvantages,we put forward the block entropy classification based weighted threshold group sparsity reconstruction algorithm(BEC-WT-GSR).In BEC-WT-GSR,the moving blocks are classified into moving texture ones and moving smooth ones,and the weighted soft threshold is used in sparsing process for the moving texture blocks.In the meantime,an exact similar block matching scheme based on Renyi entropy image segmentation is put forward.Compared with BC-ATA-GSR,experimental results show that BEC-WT-GSR improves reconstruction quality for videos with much more texture information at low sampling rates.