Fast superpixel segmentation by iterative edge refinement

Abstract

The superpixel, as an important pre-processing technique, has been successfully used in many vision applications. Introduced is a fast superpixel method called iterative edge refinement (IER). The image was first initialised as regular grids, and then concentration was on unstable pixels and relabelling them iteratively so called unstable pixels, are edge pixels around the moving boundary. It is found that the unstable pixels decrease rapidly during the iterative process, which results in a high speed-up. Experimental results on the Berkeley BSDS500 dataset show that IER achieves a segmentation performance comparable with the state-of-the-art, and moreover, runs in real-time on a single Intel i3 CPU at 2.5 GHz.