Research On Image Compression Method Based On Analysis Sparse Model Of Compressive Sensing

Modern information technology drives the rapid development of society. Imageand video signals have gradually become major carrier to obtain information in dailylife. However, with the development of technology and people’s needs, image andvideo data acquired by people is ever-increasing. Therefore, the study of image codingtechniques becomes very attractive for people.Vision sensors usually obtain image or video signal beyond the Shannon/Nyquistsampling rate of effective dimension of signal, which makes a tremendous pressurefor data storage and transmission. In the framework of compressed sensing theory,sparse representation of image signal is usually under a certain sparse base. And basedupon measurements, image can be reconstructed by optimization method for solvingsparse models. Compressive sensing theory opens up a new way for image storageand transmission.This paper mainly researches on efficient optimization reconstruction algorithmfor reconstruction model of compressive sensing. The main work includes thefollowing two parts:Firstly, TV-Wavelet-L1(TVWL1) model uses piecewise smooth characteristics ofthe image and sparse features in the wavelet domain for efficient sparse representationof image signal. This model is an efficient compressive sensing reconstruction modeland uses the partial Fourier transform for observation. However, traditional algorithmsfor solving TVWL1model often overlook synthesis/analysis methods of sparserepresentation. An image reconstruction algorithm based on the analysis sparserepresentation is proposed for TVWL1model. This algorithm divides the imagereconstruction problem into several sub-problems which are solved alternately bydifferent algorithms based upon the features of analysis sparse representation.Experimental results show that the algorithm proposed in this paper outperforms theexisting algorithms in terms of both subjective and objective quality of thereconstructed image.Secondly, Analysis-Sparse-Based algorithm is not suitable for large-scale imagereconstruction because of high space complexity. An improved algorithm is proposedto significantly reduce the space requirements in the calculation process using blockstrategy of transform matrix and conjugate gradient algorithm. Experimental results demonstrate that the proposed improved Analysis-Sparse-Base algorithm comparedwith traditional algorithms for solving TVWL1model can significantly improve thesubjective and objective quality of the reconstructed image.