Interactive Object Extraction from a Video Based on Seed-Line Tracking

This paper presents a method for extracting a non-rigid moving object from a video sequence. In order to actualize object extraction with minimal user interaction, we combine alpha matting and seed-line tracking. In the proposed method, the object region is separated from background based on seed lines given by user as a cue, and then the seed lines are automatically tracked to the next frames to actualize automatic object extraction from the frame sequence. With the alpha matting, the boundary of the object region in each frame is refined. The experimental results have shown that by the seed-line tracking user input can be significantly reduced and that the boundary refinement by alpha matting works effectively.