A new mesh based temporal-spatial segmentation for image sequence

Video segmentation, which identifies an object from an image sequence, is a crucial issue for object-based video processing. This paper presents a mesh-based silhouette segmentation technique for video sequences. We first define an active motion activity as a temporal measure, and then a new maximum a-posteriori (MAP) scheme for spatial segmentation is developed. We also introduce clique potentials for the Gibbs distribution such that the edge connectivity and the smoothness of the homogenous patches are guaranteed. Simulation results indicate that the new method achieves very good segmentation results with a much lower computational cost, as compared to existing techniques.