Motion image segmentation using 3-D watershed algorithm
Toshiyuki Yoshida, T. Shimosato · 2002
This paper proposes a technique for motion image segmentation based on object tracking. In a group of frames of a given image sequence, a single frame is first selected and a target object in the frame is segmented manually or semiautomatically with user interaction. By using the segmented frame as an initial template, the proposed technique splits the rest of the frames into object and background regions to extract the target object in the image sequence. A 3D watershed algorithm which is an extended version of the conventional 2D one is utilized as a fundamental technique. Because the 3D watershed algorithm grows starting (seed) regions not only in the spatial but also in the temporal directions, a group of frames can be segmented at a time. One of advantages of our approach is that it can keep tracking of nonrigid moving objects with rapid motion. Efficiency of the proposed technique is demonstrated by some experimental results.