Research On Image Inpainting Based On Sparse Representation

Digital image compression, transmission, storage and transformation, usually suffer from equipment influence, external interference, human destruction and natural effects that result in missing information in the images. Image inpainting technology, as an important topic in the field of digital image processing, has obtained lots of attentions and investigations in recent years. The inpainting technology can learn from the existing information, utilize a variety of algorithms and mathematical models to estimate, predict and reconstruct the missing parts of the images, aiming at having the inpainted results as close to the original images as possible. Image inpainting has been widely used in the applications of image restoration, image compression, transmission, target object removal, disocclusion, loss concealment, texture layer restoration and image edition.This paper first introduced the research background, the current image inpainting methods and mathematical models. Then we focused on the examplar-based inpainting model and patch synthesis theory. On this basis, we proposed a novel image inpainting algorithm based on the patch sparse representation, under the research of the signal sparse representation theory and matching pursuit algorithm. Then we designed C++ program to achieve this algorithm. In the experiment section, we proved the convergence of the residuals of the greedy approximation algorithm. Also, we compared the proposed algorithm to other inpainting methods such as TV, CDD and Criminisi's on the objective evaluation index. Experimental results showed that the proposed algorithm of patch sparse representation can effectively overcome texture garbage and blur. The inpaited images are visually pleasing and physically plausible.