The Research of Video Object Segmentation Based on Shape Confidence

This paper advanced a kind of video object segmentation arithmetic. This arithmetic first segmented the video reference frame by Graph cuts with user interactive behaviors, got accurate object contour. Then, it calculated shape confidence with geodesic distance, and combined color model with shape model adaptively. Finally, it constructed Graph energy function, and segments with Graph Cuts. This arithmetic can provide satisfied segmentation result when there is very little difference between foreground and background.