Geometric determinants of shape segmentation: Tests using segment identification
Elias H. Cohen and Manish Singh
Vision Research (2007)


The geometric determinants of shape decomposition were studied using a performance-based method. Observers’ identification of contour segments was shown to be systematically modulated by their curvature properties, and by the geometric properties of the enclosed region. Specifically, negative minima of contour curvature provided the best segment boundaries. Segments with negative-minima boundaries were identified with greater accuracy than those with positive maxima or inflection boundaries of comparable length. Additionally, segment identification was shown to be determined by contour length, the turning angle at part boundaries, and the width at the part’s base (hence the part’s protrusion). The results indicate that part decomposition is an automatic process. Moreover, this process is graded, i.e. parts are more strongly segmented, or more likely to be perceived, according to the strength of many geometric determinants.