Edge Grouping for Detecting Salient Boundaries with Sharp Corners

The Gestalt law of continuity, or smoothness, is an important factor in guiding the edge grouping. In practice, however, the boundaries of many salient structures of interest may not be smooth everywhere and may contain some sharp corners. This paper presents a globally optimal edge-grouping method that can detect perceptually salient boundaries with sharp corners. In this method, a set of straight-line fragments are first detected from an image by classical edge detection and line approximation. Then a modified ratio-contour method is applied to detect the most salient closed boundary by identifying and connecting a subset of these line fragments into a closed curve. The major contribution comes from the introduction of corner information into the boundarysaliency definition. The corner information is derived from the image by the widely used Harris corner detector. Experiments on some real images show that the proposed method performs better than the original ratio-contour method, when the boundaries of the salient structures of interest in the image contain some sharp corners.