Image Compression Coding Based On Signal Sparse Decomposition

Sparse representation is a tool of signal and image processing developed in recentyears, signal and image can be decomposed into a very simple representation. Imagecompression techniques play an important role in the image processing, which are widelyused in various areas of image processing. Image compression techniques currently aregenerally based on orthogonal transformation approach, but at low bit rate in this way oftendo not get the desired result of reconstruction. The signal sparse decomposition usethe techniques of over complete atom library break the orthogonal of bases, so introducing thesignal sparse decomposition ideas into the low bit rate image compression field.(1) Greedy algorithm is used to solve the problem of sparse representation, matchingpursuit algorithm is one of relatively simple greedy algorithm, but is still a large amount ofcomputation for the image sparse decomposition, differential evolution algorithm is a class ofheuristic global optimizing search method based on population, with the advantages ofdiversity of the population, robust, implicit parallelism, strong capability of global search etc.In this paper，a new optimization algorithm for image sparse decomposition is proposed bycombining greedy algorithm with the differential evolution algorithm. The proposedalgorithm used differential evolution algorithm to select the best atoms from the redundantdictionary in each step of greedy algorithm, and then achieved the optimumsparse decomposition and reconstruction of the image. The experiment results show that thequality of the image reconstructed by the proposed algorithm is improved significantlycompared with the algorithm based on the genetic algorithm.(2) Image compression encoding based on sparse decomposition is just encodingthe atom parameters and the corresponding projection coefficients, the larger proportion of thestream is the information of the location of atoms and projection coefficients of atoms, andtherefore make effective quantization coding with the two main part is the key of compression.The paper is combination the advantages of the scheme of encoding based on the sequence ofposition of atom with the scheme of encoding based on the order of projection coefficient ofatom. Using the way of based on block, utilizating rate distortion optimization criterion tofind the optimal block size, within each block coding based on the order of projectioncoefficients, and between block coding based on the sequence of the location. Experimentalresults show that the peak signal noise ratio of the reconstructed image of the propoesdmethod is better than the scheme of based on the order of atom projection coefficient underthe same bit rate. Finally, comparing the proposed scheme of image compression based onsparse decomposition with JPEG2000, the experiment results show that the proposedmethod is superior than JPEG2000at low bit rate.