Image Compression Sensing Reconstruction Based On Subband Adaptive Sparse Representation

The traditional Nyquist sampling theory states that in order to ensure the distortion-free reconstruction of the sampled signal,the frequency of the sampled signal must not be less than twice the highest frequency of the signal,and it will not only lead to increase the cost of data sampling,but also cause the subsequent data storage and transmission problems.As a new sampling theory,compressive sensing theory breaks through the limitations of the Nyquist sampling theorem and can reconstruct signals at lower sampling rates.Image reconstruction under compressive sensing is widely used in the fields of medical imaging,remote sensing imaging and visual sensing network.The basic theory of compressive sensing includes sparse representation,measurement matrix design,and nonlinear reconstruction.How to achieve efficient and accurate reconstruction of signals has always been one of the core issues in the field of compressive sensing.In view of the need of compressed imaging applications,this paper focuses on high quality reconstruction of compressive sensing images.Especially,we discuss the reconstruction method of compressive sensing based on image nonlocal similarity theory.The main work and contribution of the paper are as follows:1)Research on the image compressive sensing reconstruction method based on adaptive sparse representation of image block subband.The block-based image processing has the characteristics of lower complexity.Considering the difference of statistical characteristics of different image blocks,a compressive sensing reconstruction model based on image block subband adaptive normalization is proposed.First,the mean and standard deviation of the subband coefficients of the image blocks are estimated by the non-locally similar image block.Then,the subband coefficients of the image blocks are removed the mean and normalized based on standard deviation.Finally,the reconstructed image can be achieved by regularing the norm of the image blocks' normalized coefficients.Due to the normalized sparse representation more reasonably reflect the importance of sparse coefficients,the reconstructed image can better retain details such as texture and edge.2)Research on the image compressive sensing reconstruction method based on structure group subband adaptive sparse representation.Due to natural images have a large number of similar structural information,the structure group which is composed of similar image blocks has good sparsity.Combined with the subband coefficient normalization,the compressive sensing reconstruction model based on adaptive sparse representation of the structure group is proposed.First,the mean and variance of the subband coefficients of the image blocks are estimated by the non-locally similar image block groups.Then,it is used to normalize the subband coefficients of the structure groups.Finally,the reconstructed image can be achieved by regularing the norm of the structure groups' normalized coefficients.Because of the sparse representation of structure groups can improve the ability of sparse representation,the quality of the reconstructed image can be improved better.Objective quality and subjective visual perception are used to evaluate the quality of reconstruct image.A large number of experimental results verify the effectiveness of our algorithm in improving the quality of compressive sensing image reconstruction.