Contour-based motion segmentation using few priors

In this paper, we propose a novel method of motion segmentation through image sequence. In our work, no assumptions of boundary densities, color or textures are made about the background or the object. Neither do we assume any special motion model to the background or the object. With the help of classic Bayesian theory, an energy function describing the piece-wise smooth motion field of the image sequences are formulated. By minimizing this energy Junction, a spatiotemporal contour evolves until it enfolds the volume generated by moving objects. Very promising experimental results are provided using numerous real image sequences with natural objects and camera motion.