نبذة مختصرة : In this thesis we tackle two problems which deal with filling-in the information in a region of an image or a video, where the data is either missing or has to be replaced. These problems have applications in the context of image and video editing. The first is image inpainting, and aims at obtaining a visually plausible completion in a region in which data is missing due to damage or occlusion. The second problem concerns the propagation of an editing performed by a user in one or two reference frames of a video, throughout the rest of the video. Both problems are of theoretical interest since their analysis involves an understanding of the self-similarity in natural images and videos. At a high level, the common theme in both problems, is the exploitation and imposition of a model of redundancy (or self-similarity) to fill-in missing parts of a signal.
Processing Request
No Comments.