Motion Segmentation Using Divisive Graph Cuts

In this paper, we present a graph cut-based motion segmentation method that takes occlusion into account. We formulate the motion segmentation problem in terms of energy minimization with accounting for occlusion and minimize the energy function with the divisive graph cut algorithm where multiway minimum cuts for motion segmentation are efficiently computed through the swap move and split move of binary labels. A graph cut-based motion estimation technique is employed to estimate the motion field and occlusion between consecutive frames of the motion image sequence. Based on the motion estimate, our method segments a current frame into a number of regions of similar motion by assigning a label to each pixel. The label assignment of occluded pixels, of which the motion is not defined, is determined based on a color prior. The effectiveness of our method was verified with experimental results for various real motion image sequences.