The Taub Faculty of Computer Science Events and Talks

In recent years much work has been devoted to the development of image processing algorithms using local patches. The main idea in this line of work is to impose a statistical prior on the patches of the desired image. An algorithm following this path extracts all possible patches with overlaps from the image, and operates on each separately. The more advanced algorithms exploit also interrelations between different patches in the reconstruction process.

In this work we further study the interrelations between patches, and harness it to propose a simple yet effective regularization for image restoration problems. Our approach builds on the classic Maximum a posteriori probability (MAP) estimator, while using a novel permutation-based regularization term, following the work of Ram et. al. (2014). The permutation is obtained by a crude patch-ordering operation, and the prior employed within the MAP forces smoothness along the 1D pixel-path obtained. We demonstrate the success of the proposed scheme on a diverse set of problems: (i) severe Poisson image denoising, (ii) Gaussian image denoising, (iii) image deblurring, and (iv) single image super-resolution