A hybrid prior based general sparse image deconvolution algorithm

Compared with traditional sparse representation methods, overcomplete sparse representation is more suitable for image deconvolution. However, there have been few image deconvolution algorithms using overcomplete sparse representation. Further, among existing algorithms, a specific sparse image deconvolution algorithm corresponding to a certain sparse representation method is commonly used, which usually does not suit other methods. Therefore, in this paper, we develop a general sparse image deconvolution algorithm that can incorporate various sparse representation methods into image deconvolution depending on the applications. We propose the Bayesian framework for the presented algorithm, in which the original image is firstly modeled using a hybrid model. The statistical characteristics of the model parameters are then described using Gamma distribution. Based on the prior distributions of the original image and model parameters, we use evidence analysis method to estimate the optimal original image. The experimental results demonstrate the efficiency and competitive performance of the proposed algorithm compared with state-of-the-art algorithms.