On Non-blind Image Restoration

In this paper we develop a new fast non-blind image restoration algorithm with the goals of simplicity, high performance and computational efficiency. The speed advantage over previous algorithms, which is up to two orders of magnitude over existing schemes, is achieved by a novel choice of ground truth prior which enables the use of frequency domain methods for constrained deconvolution. In addition, we study recent likelihood models used in image restoration to guide the most effective likelihood model for image restoration. We show that the use of image derivatives in the likelihood function formulation proposed by some researchers does not lead to significant performance improvements over the standard likelihood function