Study On Compressed Sensing Based On Image Neighborhood Structural Characteristics

Compressed Sensing is a new technique for simultaneous data sampling andcompression. It breaks through the limits to Nyquist sampling theorem which needs verywide signal bandwidth when sampling. Currently, compressed sensing system used theprior that the image has sparsity in some transform domain to reconstruct the originalimage. This paper performs the researches on the image reconstruction based on newalgorithm after learning the present arts, mainly including the following three aspects.First of all, Using block compressed sensing rather than random sampling applied tothe entire image because of fast compressed sensing reconstruction. When the image isrepresented by one basis, it can’t capture the image structure effectively. A new imagecompressed sensing algorithm based on block compressed sensing and combined sparseimage representation is proposed. Simulation results show that the performance of theproposed algorithm has improvement in texture image.Secondly, the nonlocal similarity was used in image compressed sensing andcombined with the sparsity as prior. Hence, the neighborhood structure information of theimage pixels and the similarity of images are fully used. On the basis of the nonlocalsimilarity prior and the image has sparsity in some transform domain, a new imagecompressed sensing algorithm based on nonlocal similarity and alternating iterativeoptimization algorithm is proposed. The proposed algorithm solved the imagecompressed sensing problem by dealing with the following two optimization problemsalternatively: sparsity optimization problem and the nonlocal similarity optimizationproblem. And the two optimization problems are solved respectively by the iterativethresholding algorithm and nonlocal total variation.At last, the block compressed sensing provide very quick image recovery, but it istypically a reduced quality of image reconstruction.So a multiscale block compressedsensing algorithm was used to solve the problem. The image local correlation which isrepresented by autoregressive model and image nonlocal similarity was used in imagecompressed sensing as prior. Hence, the neighborhood structure information of the image pixels and the similarity of images are fully used. And then a new multiscale blockcompressed sensing algorithm based on autoregressive model and nonlocal similarity isproposed. Simulation results show that the performance of the proposed algorithm hassignificant performance improvement in visual quality of the reconstructed image andpeak signal-to-noise ratio.