Supervoxel using random walks

In this paper, we present a supervoxel generation algorithm based on partially absorbing random walks to get more accurate supervoxels in these regions. A novel spatial-temporal framework is introduced by making full use of the appearance features and motion cues, which effectively exploits the temporal consistency in the video sequence. Moreover, we build a new Laplacian optimization structure with two adjacent frames, which makes our approach to be a more efficient algorithm. Experimental results demonstrate that our method achieves better performance compared to the state-of-the-art supervoxel algorithms.