Abstract
When a scene is recorded with a camcorder, there may be disruptions such as object interference or battery exhaustion resulting in missing segments in the recorded video. A temporal video completion method that inserts a video segment captured by another camera in the place of such a faulty segment is proposed. In the proposed method, a user initially uses an interface to select both faulty and preferred segments from videos captured by two different cameras. To minimize the visual difference at the transition points between the two videos, the most similar frames to those previous to and following the faulty segment are detected around both ends of the preferred segment as the start and the end frames of the inserted segment. Note that, in order to prevent artifacts as much as possible, the length of the inserted segment can be different from the length of the faulty segment. Visually plausible video completion is then achieved by smoothing the motion and color gaps between the transition frames and completing the missing regions by using spatio-temporal features of the video. Experiments were conducted to evaluate the visual plausibility of the completed videos by changing the distance between the camera and the recorded target or between two cameras.
We apologize here, we carried 1,638 pages of color photos of the October issue article (no. 10, vol.62, pp.1633 - 1640) with monochrome. This is the correct version.
