A multi-cue information based approach to contour detection by utilizing superpixel segmentation

Abstract

Contour detection forms one of the primitive, yet inherent operations of computer vision systems. Owing to the significance of this fundamental task, a number of approaches have been proposed till date. This paper characterizes the functionality of a multi-scale feature-based edge detection strategy that exploits joint information from different feature-channels, modelled over a measure of spacial dispersion associated with structured discontinuities in an image. The issue of eliminating false edges is achieved by incorporating an iterative clustering procedure that divides the image into disjoint groups of perceptually semantic regions by constructing naturally adaptive region borders, thereby recovering precise object boundaries. From the experiments conducted on the BSDS300 dataset, it appears that the proposed detector achieves noteworthy performance by attaining promising detection results when compared to the state-of-the-art edge detection approaches.