Moving object selection based on an active curve approach

Recently, many innovating image segmentation approaches came to deal with the extraction of meaningful objects. The majority of these approaches need some user hints, such as scribbles designating the object to extract. This help to avoid a semantic gap. In the case of video object segmentation, drawing scribbles on each frame is a very long process. Scribbles drawn on one frame must be propagated automatically in the remainder of the video to make object extraction possible. In this paper, a new framework for interactive video object selection is described. This framework is built upon active curve based scribbles propagation. Scribbles drawn by the user on the first frame are considered as curves. These curves are propagated automatically by our method on the next frames to designate the same object as the one designated by the user. A spatiotemporal optimization is introduced to ensure a coherent propagation.