Morphological component decomposition combined with compressed sensing for image compression

Abstract

Basing on the fact that the cartoon and texture in one image have different morphological characteristics, we propose a new method to compress image. Combining RDWT, the dictionary sparsely representing the cartoon, and WAT, the dictionary sparsely representing the texture, the presented model can effectively obtain the cartoon and texture. Then, we reconstruct the compressed cartoon by the combination of Contourlet Transform and Compressed Sensing (CS) and reconstruct the compressed texture by the combination of single layer discrete wavelet transform (SL-DWT) and Compressed Sensing (CS). The reconstructed image will be obtained by superposing the compressed cartoon and texture. As the experimental results show, the new method has good performances for preserving large scale structure and mainly details under the low sampling rate, and ensuring the cartoon completion and texture clear. Moreover, it has higher compression rates.