Image Restoration Based On Weighting Patch Match

With the development of computer technology and the popularity of digital devices, the intelligent manipulation and editing technique of images has attracted many researchers, and has become a research hotspot. Image restoration is to restore the damaged areas or remove the occluding objects with known information, eventually acquires a visually satisfactory result, and it is an important content of image manipulation. Image restoration has been widely used in the conservation of cultural relics, film special effects, virtual reality, removal of redundant objects etc.So far, there are two main restoration algorithms: image inpainting based on the Poisson equation and exemplar-based algorithm. The image inpainting techniques fill holes in images by propagating linear structure into the target region, which is similar to diffusion procedure. This method can acquire smooth results, but fail to restore a wide range of texture regions, and the computation cost is high. Exemplar-based algorithm is a constrained texture synthesis method, which propagate the information of known region into the unknown area with a certain filling order. This algorithm is popular for its intuitive idea and efficiency. This paper proposes a novel method for computing priority based on the gradient, weighting patch match criteria, introduces a searching strategy for the best patch match that combines local search with the user intervention. On the basis of these algorithms, we present a new patch matching algorithm—restoration algorithm based on the weighting patch math. A number of experimental results demonstrate that our algorithm runs fast and well restore the images with textures and structures, obtaining visually satisfactory results.This paper firstly describes two main image restoration algorithms: image inpainting and exemplar-based algorithm. Secondly, we introduce a restoration technique that fills structure regions at a prior order, clarify the importance of the structure in image restoration, and discuss restoration results completed by various size patches and search regions. Finally, we give the weighting patch matching algorithm.