Image Inpainting Using Layer Decomposition And Partial Differential Equation

Images can be degraded,damaged,and even vandalized.This work deals with the problems of repairing damaged images using a technique called "image inpainting" since traditional image restoration,reconstruction,and enhancement can hardly be used to deal with some of the problems.Image inpainting is to use computer and specialized algorithms to automatically do the job of picture retouching that would have been done manually by an art professional in the old days.In recent years,many techniques for image inpainting have been developed and applied for various applications such as removing scratches,defects,super-imposed text,or even entire object from the scene in an undetectable manner.However,as it is a new field of research,many problems remain unsolved,therefore extensive investigations are needed to advance the research and make the techniques practical.In this work,we propose several new methods in order to contribute to the development of this attractive and useful area in image processing.First,we propose a method to improve quality of photographs taken under poor weather conditions.We use a thresholding operation to segment the image to obtain a template for the sky.The gloomy sky is replaced with a patch,which is cloned from an available picture of blue sky,using an algebraic operation.For the fine retouch of complicated boundaries where transition between the sky and ground objects is not clearly defined,a texture growing method is introduced to generate the texture along the boundary.This makes the modified image look natural.Second,we develop a layered image inpainting method.The damaged image is first decomposed into three layers:cartoon,edge,and texture using DCT.The cartoon and edge layers are repaired using an adaptive offset operator that can fill-in damaged image blocks while preserving sharpness of edges.The missing information in the texture layer is generated with a texture synthesis method.Compared with some existing techniques,the proposed method can produce good image with less blurring of the edges and a faster processing speed.Third,we use a heat transfer model to solve the problem of image inpainting,in which the pixel gray level is made analogous to temperature.A finite difference method is developed to solve the partial differential equations(PDE) numerically,using double-grid and single-grid iteration schemes.Information contained in the neighborhood is propagated into the missing area progressively.This method is relatively simple as compared to other PDE approaches,and gives satisfactory fill-in effects of the cartoon layer of the damaged image.Texture information is extracted from the area around the hole,and added to the repaired patches to produce a more natural appearance.Experiments have been carded out on images with various types of damages to show the effectiveness of the proposed methods.