Research Of Image Inpainting Models Based On PDE

Digital image inpainting plays an important role in image processing, including inpainting destroyed images, disocclusion, image compression and so on. The paper focuses on research of image inpainting models based on PDE.About the initialization of destroyed area, we take the relativity of the destroyed area and the natural area into consideration, and then the "related random initialization" is proposed. The experimental results show that compare with previous schemes, the method can make the inpainting outputs more natural and improve the inpainting efficiency as well.About the inpainting models, we mainly observe the two important characters of images, the gradient and the curvature of the isophotes, and introduce the G-C model based on total variation. The model can restore the information of the destroyed area naturally and connect the broken level sets smoothly. Meanwhile, we combine the diffusion along the tangent and normal direction, and then present a combined model based on G-C model and BSCB model. The results show that the novel model performs better than G-C model in inpainting efficiency.