Datensatz

Zusammenfassung

During the last years, many image processing and manipulation techniques have evolved and have become more and more popular, reaching from scienti c theory of mathematical image processing to general-purpose software for artistic image editing, e.g. Adobe Photoshop [Ado06] or GIMP [GIM06]. These tools are adequate for many tasks, but require a high amount of user interaction. In this work, we focus on the image inpainting problem: given an image, an undesired object is to be removed from the image, and the "hole" arising is lled in a visually plausible way. To this end, we extend existing work on this topic, and apply the basic idea to a higher-dimensional problem: video inpainting. Here, an undesired object is to be removed from a video sequence. As a video is nothing else than a sequence of images in time, this requires a method producing consistently lled holes over time. Compared to many existing approaches for image and video inpainting, we aim at a method with as little user interaction as possible. The minimum interaction required for our method is the selection of the region to be inpainted.