Blind deconvolution of natural images using segmentation based CMA

In this paper, we analyze the applicability of Constant Modulus Algorithm (CMA), one of the most widely used and tested blind equalization technique to blind image deconvolution. With a detailed mathematical analysis, we show that the strong correlation between the neighboring spatial locations found in natural images becomes a major constraint on the convergence of CMA. In order to overcome this constraint, we introduce a novel image pixel correlation model in relation with natural image statistics. Based on this model, a segmented blind image deconvolution through CMA is proposed. The robustness of the proposed algorithm with natural images is discussed in terms of efficiency and effectiveness.