Image Colorization And Image Inpainting Based On Sparse Representation And Dictionary Training

Image colorization has been an active and challenging research topic. Initially color-ization was defined as an assistant processing technique using computer, with time go-ing, the research in the field of colorization has more and more important value. On the one hand it can solve the problem of how to give the old pictures and movies color, on the other hand when we analyze the data of pictures and videos the color information has great research value, so the research of colorization has more and more extensive application in the field of medical, movie, space exploration and other fields in science and industry.The technology of Image inpainting is to repair the damaged part in the image with a few rules using the efficacious information. It must make sure that the repaired image has similar visual effect with the primary image as much as possible. The technology has played an increasingly significant role in the field of special effect of film, error concealment of video image, art restoration. Now it has become a research hotspot in the field of image processing.In this paper, our theoretical basis is the theory of sparse representation, we make a deep study in the research of how to represent the image signal with sparse representa-tion. With focusing on the method of dictionary training, we do an exploratory and in-novative research in the field of image colorization and image inpainting. Finally we propose two algorithms based on the research:the algorithm of gray image colorization and the algorithm of image inpainting.In the algorithm of gray image colorization, image segmentation is not necessary, the colorization is made on the entire image, so it is a global and automatic colorization algorithm. Firstly, a joint dictionary is trained by reference color images according to the correlation among the luminance, feature and color of trained images. And then the sparse coefficients under the joint dictionary for the grayscale image are computed by using its luminance and feature information. Finally, the color information is recon-structed using the above joint dictionary and the obtained sparse coefficients. Experi-mental results demonstrate that the algorithm is effective and efficient, especially for those monotonous images.In the algorithm of image inpainting, it uses a large number of samples and the effi-cacious information in the damaged image to repair the image. We construct a diction- ary, then with the purpose of making the feature information of the sample images and the primary data information have the same sparse representation under the dictionary we train the dictionary. Final we restore the information of the damaged region in the image using the common coefficient and the trained dictionary. Experimental results demonstrate that the algorithm is effective and efficient, especially when the reparation is to fill up the damaged area in the image.