Research On Image Reconstruction And Denoising Method Based On Sparse Representation

The demand for image information is on the increase in world market.In the process of information sampling and transmission,the traditional Nyquist theorem brought about by the waste of sampling resources,expensive hardware,low efficiency of information processing and other limitations become increasingly prominent.The premise of signal sparse property of compressed sensing sampling coding technology,which avoids the generation of redundant data,effectively improves the efficiency of signal processing and reduces the requirement of sampling rate.Saving data storage space and transmission costs.Based on compressed sensing theory,this paper deeply studies the image reconstruction and denoising methods based on sparse representation,and uses the sparse representation model for image transmission,coding,and denoising of contaminated images in order to recover efficiently and accurately restore the complete image signal.The paper has made the research achievements in the following aspects.(1)The third chapter puts forward a patch group weight encoding method,to solve the shortage problem of mixed noise removal.First of all,by using nonlocal similar patches to extract patch groups from training images,and then using patch group to learn the nonlocal self-similarity prior.In the end,integrate the sparsity prior model and nonlocal self-similarity prior model into regularization term and encoding framework.Experimental results show that the method that proposed achieves significant improvements over the previous sparse reconstructed image methods,the PSNR improved of 0.036 ~ 2.865 dB,and obtains better quality of denoising image.(2)The fourth chapter puts forward a nonlocal PCA based clustering the sparse representation,to solve the shortages problem of corrupt image restoration.First of all,obtain sparse coefficient value by image nonlocal self-similarity,and then centralize the sparse coding coefficients of the observed image to sparse coefficient value.In the end,make the sparse coding coefficients of the degraded image as close as possible to those of the unknown original image through learning the dictionary.Experimental results show that the method that proposed achieves significant improvements over the previous sparse reconstructed image methods,the PSNR average raised 0.5653 dB,and obtains better quality of image restoration.(3)The fifth chapter puts forward a Gaussian scale patch group sparse representation method,to solve the shortage problem of image restoration.First,utilized the nonlocal similar patches to extract the patch groups,and then using the simultaneous sparse coding to develop a nonlocal extension of Gaussian scale mixture model.In the end,integrate the patch group model and Gaussian scale sparsity model into encoding framework.Experimental results have shown that the proposed method can both preserve the sharpness of edges and suppress undesirable artifacts.And often delivers reconstructed images with higher subjective/objective qualities than othercompeting approaches.The PSNR increased of 0.02 ~ 0.64 dB.