Image Compressive-Sensed Coding And Recovery Using Self-Similarity Contexts

Image compressive sensing states that the image having sparse or compressive representation can be exactly reconstructed by using incoherent measurements acquired at the sub-Nyquist rate,and it has established the theoretical foundation for low-cost image and coding.Image compressive sensing is widely used in medical imaging,low-complexity coding,radar imaging,optics,video surveillance and other fields.Therefore,it has attracted extensive attention from academia and industry.In order to solve the problems of heavy storage burden and high computational complexity of restoration,image compressive sensing often implements block measurement and restoration.However,existing works fail to adaptively measure according to the sparsity of image blocks,resulting in a waste of measurement resources and serious blocking artifacts in the restored image.Although image features can be used to measure image sparsity,commonly used features,like variance,edge,and saliency,only consider the internal structure information of image patches,and ignore the correlation between image patches.More seriously,when the original pixels are unknown,these image features are difficult to extract.To solve the above problems,this paper first uses the compressive sensing measurements to extract context features to measure the correlation between patches in the measurement domain and reflect the variation of the sparsity of image patches.Then,on this basis,a context-directed image compressed sensing coding and restoration method is proposed.The main research contents of this paper are as follows:(1)An algorithm for extracting context feature from measurement domain is proposed.For the problem of unknown original pixels in compressive sensing system,an algorithm of extracting context feature from the measurement domain was proposed.The algorithm firstly calculated the similarity weight between the central block and the surrounding non-overlapping blocks based on the measurements of the blocks.Then,the correlation surface about the center block was constructed by using all the similarity weights.Finally,the context feature of the center block was calculated by calculating the average of the correlation surface composed of all the similar weights.The experimental results show that the context feature extracted from measurement domain can better reflect the sparsity variation of image,and demonstrate comparable performance with the context feature extracted from pixel domain.(2)A self-similarity context-directed image compressive sensing coding scheme is proposed.The scheme extracts context feature according to the measurement domain to guide the adaptive measurement,avoiding the redundancy or deficiency of the measurements,and proposes zero-padding predictive quantization to overcome the problem of inconsistent measurement vector length caused by adaptive measurement.The experimental results show that the proposed self-similarity context-directed image compressive sensing coding scheme effectively suppress the blocking artifacts in the restored image and has better rate-distortion performance.(3)A self-similarity context-directed image compressive sensing restoration algorithm is proposed.The algorithm firstly introduced a probabilistic framework to model the restoration problem based on the Bayesian model.Then,the implicit correlation between the context feature and the parameters of the restoration model was deduced,and the parameters were predicted.Finally,an adaptive restoration scheme was constructed by adaptively tracking the sparsity variation of the blocks.The experimental results show that,compared with the traditional iterative restoration algorithm,the proposed self-similarity context-directed image compressive sensing restoration algorithm significantly improves the visual quality of the restored image and ensures a lower computational complexity.