Layered Image Inpainting Based On Sparsity

Image inpainting is a process that gains the unknown information of a damaged image from taking advantages of the known information. Due to improper storage, man-made damage reasons, some artworks and photoes may hold the problems like scratches, degradation, and so on. If we process broken image only, it may take lots of time, even we can't make sure the result is appropriate. However, digital image inpainting technology is a better solution to this problem. The process can be completed by computer automaticly. Now, digital image inpainting has played a great role in image super resolution, image compressing, error concealment, video transfer, and so on.This dissertation has mainly studied image inpainting algorithm based on sparse decomposition. Combines image inpainting based on sparsity and image decomposition model, it designs the layered image inpainting algorithm based on sparsity. The main work shows as follow:1. The theory of image inpainting based on sparse decomposition is analysed. The MP algorithm is applied to image inpainting, and some simulation experiments are made. Because the MP algorithm takes a lot of time, here the dissertation selectively researches the image inpainting algorithm based on sparsity. Its speed is improved obviously. Based on sparsity theory, in refrence to the characteristics of structure and texture information in images, here the method of layered inpainting and superposition processing are brought to mind.2. The image decompotion models are researched, such as DCT, ROF and VO. This dissertation analyses the principle and characteristic of the three models, it also compare the decomposition result of image for three models. Experiments shows that VO model is more effective on image decomposition than the other two.3. In view of sparsity, the layered image inpainting algorithm and superposition processing are researched. Image is decomposed into two layers:structure and texture, choosing the Curvelet, DCT represent the images respectively, the last result has got through overlapping the two resulted images. The experiment has demonstrated that DCT model is not effective on layered image inpaiting. For the next step, more research are made on layered image inpaiting based on VO model and superposton processing. Furthermore, it has provided some examples verifies the feasibility of this method. Superposition processing integrates the strongpoint of solo transform Curvelet and DCT, although some blemish occurs, the integral effect of the image has improved to some degrees.