Convexity grouping of salient contours

Abstract

Convexity represents an important principle of grouping in visual perceptual organization. This paper presents a new technique for contour grouping based on convexity and has the following two properties. Firstly it finds groupings that form contours of high convexity which are not strictly convex. Secondly it finds groupings that form both open and closed contours of high convexity. The authors are unaware of any existing technique which exhibits either of these properties. Contour grouping is posed as the problem of finding minimum cost paths in a graph. The proposed method is evaluated against two highly cited benchmark methods which find strictly convex contours. Both qualitative and quantitative results on natural images demonstrate the proposed method significantly outperforms both benchmark methods. © 2011 Springer-Verlag Berlin Heidelberg.