| Image inpainting, as an important research point in image processing, is attracting more and more researchers in areas of computer vision and computer graphics. The partial differential equations(PDE) is one of the mainstream methods of image processing, image inpainting models based on partial differential equations have good results on non-texture image. TV(Total Variation)-L1 model has been successfully used to gray image. However, this model cannot be directly applied to color texture image due to coupling of different layers in color images.In order to restore the large broken texture color images, we extend TV-L1 model to non-local CTV-L1 model. The main work and innovations include the following: first, several classic image inpainting models were briefly introduced. Mainly focuses on the micro simulation BSCB(Bertalmio, Sapiro, Caselles, Balleste) model, TV(Total Variation) model and CDD(Curvature-Driven Diffusions) model, then analyse advantages and disadvantages of the above models. Second, we studied and analysed the non-local TV-L1 model, including its principle, methods and applications. Third, due to non-local TV-L1 model cannot be directly used on color images inpainting for the coupling of different layers in color images, we proposed a non-local CTV-L1 model for color images inpainting. Technically, this new model includes both non-local operators that is capable of inpainting large broken texture image and CTV(Color Total Variation) regularization term that is able to preserve the edge of color image. In addition, L1 data term, which can preserve the contrast of inpainted images and also can remove anomalies and irregularities from images. By introducing auxiliary variables and Lagrange multiplier, we design the simple augmented Lagrangian algorithm for this model to improve its computational efficiency. Finally, some numerical experiments are presented to validate the performance of the model and algorithm proposed in this paper. |
PDF Full Text Request