Streaming Video Segmentation via Short-Term Hierarchical Segmentation and Frame-by-Frame Markov Random Field Optimization



An online video segmentation algorithm, based on short-term hierarchical segmentation (STHS) and frame-by-frame Markov random field (MRF) optimization, is proposed in this work. We develop the STHS technique, which generates initial segments by sliding a short window of frames. In STHS, we apply spatial agglomerative clustering to each frame, and then adopt inter-frame bipartite graph matching to construct initial segments. Then, we partition each frame into final segments, by minimizing an MRF energy function composed of unary and pairwise costs. We compute the unary cost using the STHS initial segments and the segmentation result at the previous frame. We set the pairwise cost to encourage similar nodes to have the same segment label. Experimental results on a video segmentation benchmark dataset, VSB100, demonstrate that the proposed algorithm outperforms state-of-the-art online video segmentation techniques significantly.