Nonlocal regularization on weighted patches for image deconvolution

Regularization-based methods have been found widespread application in image deconvolution. Local regularization approaches have made outstanding performance for edge maintain, such as total variation regularization. However, they have weakness in textures preserving. To handle images of hybrid edges and textures, we propose an iterative regularization method that utilizes weighted patches along with nonlocal means filtering. The weighted patches allow us to preserve textures as well as edges. The iterative form in the regularization provides improvement in detail restoration. This nonlocal regularization method reveals favorable recovery of structures in images, especially for textures and edges. Experiment results confirm such superiority with comparison to local regularization methods.