Blind image separation based on reorganization of block DCT

Blind image separation consists in processing a set of observed mixed images to separate them into a set of original components. Most of the current blind separation methods assume that the sources are as statistically independent or sparsity as possible given the observations. However, these hypotheses do not hold in real world situation. Considering that the images do not satisfy the independent and sparsity conditions, so the mixed images cannot be separated with independent component analysis and sparse component analysis directly. In this paper, a method based on reorganization of blocked discrete cosine transform (RBDCT) is first proposed to separate the mixed images. Firstly, we get the sparse blocks through RBDCT, and then select the sparsest block adaptively by linear strength in which the mixing matrix can be estimated by clustering methods. In addition, a theoretical result about the linearity of the RBDCT is proved. The effectiveness of the proposed approach is demonstrated by several numerical experiments and compared the results with other classical blind image methods.