Image Inpainting Algorithm Research Based On Partial Differential Equation

Since Sichuan is on the Europe and Asia earthquake belt, so there are many earthquakes every year. The earthquake brought enormous loss of personnel and property, so how to start the first time rescue after the earthquake has become a very important problem! So seismological bureau's of Sichuan starts a project, it names"Sichuan disaster receiving and processing system platform", digital image of the disaster information acquisition belong this project. People will evaluate the severity of earthquake damage based on these figures, so this figures provided a powerful guide for the launch of the rescue .The digital image in the transmission process due to various disturbances, it is easy to be contaminated, it performs noise and damage etc, Decision-makers couldn't analyze the right result, so the paper starts the system research for the image inpainting.Image inpainting is a hot topic in the current computer graphics and computer vision, it is applied widely in academia and real life. In the academic, this technology can provide clear and complete image for many other areas; at the same time, in practice, it can repair the defect photo, image, remove the occluder of target object.This paper describes the classic model in pixel domain------TV model, it is a anisotropic diffusion model, which can protect image edge and details effectively, but if we still use anisotropic diffusion in the flat areas, it will make false edge which leads a poor effect. So this paper makes some progress for TV model, and propose a improved TV model. It introduced two parameters, one is semi-adaptive parameter p, another one is protection parameter of gradient scope. p is used to chose the different diffusion model in different area. It means that p value chose to 1 when the damage area contain lots of edges and details, it makes the model close to TV model which lead a anisotropic diffusion; p value chose to 0 when the damage area contain lots of flat areas, it makes the model close to thermal model which lead a isotropic diffusion. K is used to control the diffusion speed and the scope of protected gradient. The bigger K, the faster diffusion speed, but the less edge; otherwise the more slowly speed and the more widely scope of gradient.Then, this paper describes the TV-wavelet model in wavelet domain. The fact of the TV-wavelet model is to repair the wavelet coefficients in wavelet domain, but the inpainting regular is controlled by TV model which is in the pixel domain. It makes the image which has repaired the wavelet coefficients satisfy anisotropic diffusion, it means the edges and the details get the well protect. Because the TV-wavelet model is introduced by TV model to wavelet domain, and the improved TV model's effect is better than the TV model, which can chose different inpainting model in different area. So we consider that when repair wavelet coefficients in wavelet domain, this paper propose lead the improved TV model to the wavelet domain, that's means the inpainting regular of wavelet coefficients is controlled by the improved TV model, It names improved TV-wavelet model. Due to the well performance of improved TV model in the pixel domain, so the improved TV-wavelet model owns the well adaptability in the wavelet domain.The result of experiments shows that the effect of improved TV model is better than the TV model in the same number of iterations; and the effect of improved TV-wavelet model is better than the TV-wavelet model.