PROJECT TITLE :

Multi-Scale Patch-Based Image Restoration

ABSTRACT:

Many image restoration algorithms in recent times are based mostly on patch processing. The core plan is to decompose the target image into absolutely overlapping patches, restore each of them separately, and then merge the results by a lucid averaging. This concept has been demonstrated to be highly effective, leading often times to the state-of-the-art ends up in denoising, inpainting, deblurring, segmentation, and alternative applications. While the above is indeed effective, this approach has one major flaw: the prior is imposed on intermediate (patch) results, rather than on the ultimate outcome, and this is often sometimes manifested by visual artifacts. The expected patch log likelihood (EPLL) method by Zoran and Weiss was conceived for addressing this very downside. Their algorithm imposes the previous on the patches of the final image, that in turn leads to an iterative restoration of diminishing effect. In this paper, we tend to propose to more extend and improve the EPLL by considering a multi-scale prior. Our algorithm imposes the very same previous on totally different scale patches extracted from the target image. Whereas all the treated patches are of the identical size, their footprint in the destination image varies because of subsampling. Our theme involves alleviate another shortcoming existing in patch-based restoration algorithms—the fact that a local (patch-based) prior is serving as a model for a international stochastic phenomenon. We tend to motivate the use of the multi-scale EPLL by proscribing ourselves to the simple Gaussian case, comparing the aforementioned algorithms and showing a transparent advantage to the proposed technique. We have a tendency to then demonstrate our algorithm in the context of image denoising, deblurring, and super-resolution, showing an improvement in performance both visually and quantitatively.