Motion-based video segmentation with boundary refinement

Motion-based video segmentation remains an important problem in video processing. A promising approach that has received significant attention formulates the problem as an energy minimization within a MAP-MRF framework. While a great deal of progress has been made toward finding robust and computationally reasonable motion segmentation methods, automatically generating such a segmentation that performs well at motion boundaries remains a challenging task. To address this problem, we propose and demonstrate a motion-based video segmentation method that uses an automatic color-based boundary refinement strategy to obtain a boundary-accurate segmentation within a reasonable computation time.