Real-time Texture Boundary Detection from Ridges in the Standard Deviation Space

This paper presents a novel texture boundary detector called the standard deviation ridge detector. At 43.29 frames per second, it is one of the few texture boundary detectors that can run in realtime. With its Berkeley segmentation benchmark F-statistic of 0.62, the algorithm outperforms all existing realtime texture boundary detectors. The use of the boundary detector would induce noticable improvements to all realtime machine vision applications. In addition, a ridge detector, which is better suited to realtime than other approaches, is presented as part of the proposed algorithm.