Low Bit Rate Image Coding Based Overcomplete Sparse Representation

A sparse image representation algorithm, named multi-atoms matching pursuit (MAMP), and its application to still image coding are proposed. The MAMP algorithm decomposes images sparsely by selecting several atoms at each iteration. The approximation performance of the MAMP algorithm is comparable with those of matching pursuit (MP). Meanwhile, the speeds of sparse decomposition in MAMP algorithm are greatly improved. A multiscale ridge let dictionary is designed to sparsely represent images. The coding performances are tested. Experimental results show the performances of this paper outperform the state of the art JPEG-2000 in terms of both PSNR and visual quality, especially at low bit rate.