Research On Digital Image Inpainting

Image inpainting is a research hotspot of computer graphics and computer vision. It's applied widely in damaged image restoration and target objects removal. Total variation model (TV model) has a defect that can't meet the"connectivity"guidelines of human being's vision, which results in the bad effect of inpainting when the damaged area is larger. Then, a curvature diffusion coefficient is introduced into the TV model to get the curvature driven diffusion model (CDD model). The coefficient can enhance the diffusion when the curvature is large. The CDD model's inpainting effect is better than TV model because the CDD model can connect the big damaged area. But due to the curvature diffusion parameter, it makes the CDD model be a 3rd-order partial differential equation, which extends inpainting time greatly.In this paper, an adaptive image inpainting algorithm based on CDD model is proposed to overcome the problem of TV model and CDD model. An adaptive coefficient of gradient p and an adaptive coefficient of curvature q are introduced into CDD model. The adaptive coefficient of curvature q can choose different inpainting model by curvature, which makes the model choose CDD model only in large curvature to reduce inpainting time. The adaptive coefficient of gradient p can choose different diffusion way by image's degree of gray change, which makes the model choose anisotropic diffusion when damaged area contains many edges, and otherwise choose the thermal diffusion equation to obtain a better inpainting effect.The performances of the different models are analyzed by comparing with the experimental results of TV model, CDD model, improved CDD model and the proposed adaptive model. The experimental results show that the inpainting effect and time of the proposed adaptive model based on CDD model is better than the other models.