Image compressed sensing based on the similarity of image blocks

Compressed Sensing (CS) theory has recently received amount of attention in the image compression filed. The sparser the signal has, the better the performance recovery. Most wavelet-based reconstruction methods of CS are developed under the assumption that the small wavelet coefficients are close to zero. In other words, a part of image information has been lost before measurement sampling. As we know, most of images have many similar areas. In order to avoid the image information being lost as little as possible, in this paper, a new CS scheme based on the similarity of image blocks is proposed in wavelet domain. Instead of processing the image as a whole, the image is firstly divided into small image blocks. And a clustering algorithm is presented to gather the similar image blocks into a group. Experiments on images demonstrate favorable performances of the proposed method.