MovieCut — Dynamic object segmentation in image sequences

Object segmentation in videos or image sequences is a crucial task and has gained more and more attention in the past years, as computer performance is increasing. Segmentation may be carried out in the field of video surveillance with automatic object tracking or as a composing and editing task in the post-production of movies. In the first case a suitable real-time performance is wanted with a drawback in segmentation accuracy, whereas in the latter one accuracy is the most important factor. We will show a new approach for dynamic segmentation and tracking of objects in image sequences that is based upon graph-based segmentation techniques from still image segmentation. Although the current performance of our algorithm is far from being realtime, we think there is enough room left for optimization in our implementation. We demonstrate the usability of our new MovieCut algorithm on some examples and give some hints to improve the overall performance.