An Improved Image Inpainting Algorithm Based On Patch Matching

With the development of the times,artificial intelligence and other technologies have been promoted to be more and more common used.Digital image intelligent processing technology,as one of the artificial intelligence has been widely used.Image inpainting is a very important part of the image processing technology.Image inpainting is a processing that fills in missing information about damaged areas in an image.The purpose is to restore the damaged image and make it impossible for the observer to detect that the image had been defective or repaired.Whether it is in the field of digital photo processing applications,such as damaged artifacts murals,old photographs repairing,etc.or in the field of image-oriented high-level understanding of the field of researching,such as removing some object and content integrity before we do image recognition,both have a wide range of needs.Digital image restoration is a mathematically illogical inverse problem,which requires artificially a given priori knowledge of the image to guide the image restoration process to produce a visual integrity that can be accepted by the human eyes.So how to construct an effective image prior model is the key problem of digital image restoration.Today,image inpainting technology has been divided into two parts.One is PDE image inpainting for the small areas,and the other one is texture-based inpainting technology for big areas.Bertalmio firstly adopted PDE technology into image inpainting.The basic idea is to fill the damaged area according to the principle of information diffusion in physics.When the domain to be filled is small,the inpainting effect is pretty nice.But if the domain is very big,the result may be fuzzy.So the inpainting technology based on PDE isn't robust enough,therefore texture-based inpainting has gradually been more and more popular,and has attracted many people to do research on it.This kind of algorithm regardless of the size of the damaged area can achieve very good results.In this article,we mainly introduce and studied Criminisi algorithm.On the base of the Criminisi algorithm,we have prompted the priority calculation method,the searching method of the best matching block and the priority updating method of the repaired point.In the case of finding the largest sample block,the gradient term is introduced into the original Criminisi algorithm,which fully considers the surrounding index of the sample block,it makes the searching for the largest priority block more reliable and avoid the less information The structure similarity information is introduced in the process of finding the most similar sample block,and the whole information of the sample block is taken into account,rather than using the differences between each pixel point,So that the matching between the sample block and the block to be repaired more reliable information.During the research we have found that,with the inpainting going on.Criminisi algorithm in the updating of the confidence term will be a rapid decline,which makes the data items,gradient items and other information nearly has no effect.So it may have an unbelievable effect on the filling order.Focusing on all those problems,the paper adds a concept which is called the distance term.And the new method sets a threshold,while the confidence term is smaller than it,let it equal to the threshold,thus ensuring the validity of the data item and the gradient term.Finally,the simulation results are compared with Criminisi's method and others.And the effectiveness of the improved algorithm is illustrated by comparing the different images' results,including the images with rich texture,complicated structure and more curves.