Dynamic Random Walk for Superpixel Segmentation

Abstract

In this paper, we present a novel Random Walk model called Dynamic Random Walk (DRW) for superpixel segmentation. The proposed DRW adds a new type of node called dynamic node to enrich the features of labels and reduce redundant calculation. By greedily optimizing the Weighted Random Walk Entropy (WRWE), our DRW can consider the features of both seed nodes and dynamic nodes, which enhances the boundary adherence. In addition, a new seed initialization strategy, which can evenly distribute seed nodes in both 2D and 3D space, is proposed to extend our DRW for superpixel segmentation. With this strategy, our DRW can generate superpixels in only one iteration without updating seed nodes. The experiment results show that our DRW is faster than existing RW models, and better than the state-of-the-art superpixel segmentation algorithms in both efficiency and the performance.