Partial Differential Equations-based Image Repair

Image inpainting is very important in image processing, it is a hotspot in computer graphics and computer vision, it is used widely. Image inpainting is to deal with damaged image, photo and occlusion. How to inpaint the damaged image and fill some information into the occluded regions, it is an important problem. At present, there are two themes related to inpainting, one is image inpainting, the other is texture synthesis. TV inpainting model is the focus of this dissertation, we carry out research on the principle, methods and applications about image inpainting.Firstly discuss the essential theory related to image inpainting, including the principle of image inpainting, isotropic diffusion, anisotropic diffusion and evaluation criteria of image inpainting. Then introduce six types inpainting model based on Partial Differential Equation(PDE), focus on Curvature-Driven Diffusions model (CDD) and Total Variation model(TV). The CDD model is through curvature to control its diffusion speed, realizes the connectivity principle. TV model could preserve image edges effectively, denosing and inpaint narrow strip, but it could fit gray image, not fit color image. A improved channel coupled diffusion model based on TV is proposed. Through experiment, results show that the color TV model could inpaint damaged image and preserve image edges better. But it could make edge blurred for color image with noise, then color TV model is improved, experimental results show improved color TV model could maintain the clear edge of image, good results and denoising.