A Study On Example-based Image Inpainting Algorithm

Image inpainting desires to fill in the data in missing area using the information from the observed region of an image. This field of research has been active since 2000 after the image de-noise, de-blur, superresolution and texture synthesis have been increasingly matured. Image inpainting can be widely used in the numerous applications such as restoring image from scratches or text overlays, object removal, getting rid of red eyes and so on. It also can be used to remove the occlusion of objects which will significantly improve the recognition rate of objects especially in the infrared image.In this paper, we study the example-based image inpainting problems and propose a new image inpainting method under alternating direction method of multipliers(ADMM) optimization framework. Inpainting processing order of each patch will be determined by improved patch sparsity first. Then a metric measuring similarity between two patches is proposed by Laplace probability distribution of DCT tight frame system coefficients which is more robust to distinguish difference of patch than sum of squared differences by ?2 norm. The patch-based dictionary can be adaptively established by proposed metric from the nonlocal data and is similar to target patch. The results show that the proposed patch-based image inpainting method is efficient in interpolating large missing region and provides more plausible from the points of visual effect.