A hierarchical segmentation tree for superpixel-based image segmentation

In this work, we introduce the concept of the superpixel association and propose a hierarchical tree for image segmentation with superpixel associations. Different from other unsupervised image segmentation algorithms, our method can generate segmentation without needing specifiednumber of segments. We adopt a bipartite graph to fuse the segmentation clues and propose a voting method to measure the similarities between the superpixel associations. The experiments on two public image segmentation datasets show that our algorithm is competitive when compared other state-of-the-art unsupervised segmentation algorithms.