Sparse Regularized Biomedical Image Deconvolution Method Based on Dictionary Learning

摘要

Image deconvolution is a very difficult problem, the regularization image deconvolution model based on dictionary learning in the framework of sparse decomposition is established, including dictionary learning and deconvolution. Fist of all, the conditional constrained optimization problem dictionary update is transformed to unconstrained functional model in this paper. The convolution model regularization respectively defines image smoothness in the Besov and the sparsity of decomposition coefficients, which is measured by a similar potential function. Secondly, we adopt split Bregman method and introduce alternative function to solve the model, and deduce the solution process in detail, obtain the model solving algorithm and discuss the regularization parameter selection of the model. Finally, we perform deconvolution experiments with different types degraded situation images. The results show that the peak signal-to-noise ratio and improved signal-to-noise ratio of this paper model is much higher than those from model (1) and GPSR model.