Robust image de-blurring with inaccurate blur kernels

Most existing non-blind image deblurring methods assume that the blur kernel is free of error. However, it is often unavoidable in practice that the input blur kernel is erroneous to some extent. Sometimes the error could be severe, e.g, for images degraded by non-uniform motion blurring. When an inaccurate blur kernel is used as the input, significant distortions will appear in the image recovered by existing methods. In this paper, we present a novel convex minimization model that explicitly takes account of errors in the blur kernel. The resulting minimization problem can be be efficiently solved by the so-called APG method. Also, a new boundary extension scheme is incorporated in the proposed model to further improve the results. The experiments on both synthesized and real images showed the efficiency and robustness of our algorithm to both the image noise and the model error in blur kernel.