Image Inpainting And Zooming Algorithm Based On Total Variation

Digital image inpainting, which is an important field of the current image processing and computer vision research, is mainly using the information of not damaged areas to fill the lost or damaged areas, according to a certain rule, making the restored image and the original image as close as possible. At present, the image restoration techniques have been widely applied in restoring damaged photos, removing target, error concealment, image magnification and so on.Restoration techniques based on partial differential equation (PDE) and texture synthesis are current two classes of important methods in the spatial domain. They are respectively used to restore structure images and texture images, but these methods need the inpainting boundary be relatively clear. In the transmission, the lost of frequency coefficients often lead to damaged areas throughout the image, and repaired areas have not definite boundary. Thus, the effect of directly using the spatial domain method is not ideal.Firstly, the significance and study situation of image inpainting are introduced in brief. We analyze several algorithms in spatial and frequency domain, such as BSCB model, TV model, and TV model in wavelet domain.Secondly, we focus on the research of nonlocal total variation (TV) of image inpainting algorithm in the discrete cosine transform (DCT) field. This algorithm, which combines the spatial domain and frequency domain information, introduces a nonlocal total variation inpainting method. This problem is solved by means of the rapid Bregman Operator Splitting method. Experimental results show that this algorithm makes full use of global information of an image, not only keeping the geometry information, but also effectively repairing the texture information.Finally, we present using TV model realizes image magnification, by means of graph cut technology minimizing energy function. Experimental results show that the method can effectively preserve geometrical edge, getting the global optimal solution of energy function and avoiding divisor equal to zero in the traditional steepest descent method. Thus, it can inhibit ladder effect, achieving a better magnification effect.