Purchase this article with an account.
Joeri Winter, Johan Wagemans; Segmentation of object outlines into parts: From a large-scale normative study to a model. Journal of Vision 2001;1(3):421. doi: 10.1167/1.3.421.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previous studies of object segmentation have used a limited set of stimuli and a small number of subjects. We presented 88 outline versions of line drawings of everyday objects from a standardized set (Snodgrass, J. G., & Vanderwart, M., 1980, Journal of Experimental Psychology: Human Memory and Learning, 6, 174–215). Half of these were easy to identify and half difficult, according to an independent identification test. Two paper-and-pencil studies with a large number of subjects (n = 200 each) were performed to obtain good normative segmentation data. These data were then entered into the computer and analyzed in comparison with the complete curvature graphs of all the outlines. In Experiment 1, subjects drew lines that crossed the contour twice to indicate the object's parts. In Experiment 2, subjects segmented the contour itself. In both experiments, negative minima of curvature (m-) were most frequently and consistently chosen as segmentation points but other curvature singularities were marked too. In Experiment 2, positive maxima (M+) were also chosen frequently, possibly because these are salient points in themselves. In Experiment 1, a wide variety of lines was often drawn through one strong m- point and a weaker corresponding point elsewhere on the contour (e.g., inflections). In these conditions, more global shape factors such as the distance between segmentation points, collinearity of contour fragments and symmetry of the whole object or the resulting part clearly affected the segmentation. In both experiments, segmentation points were more frequent and more consistent with easy-to-identify objects than with difficult ones, suggesting a top-down influence on the segmentation process. Based on these results, we will develop an integrative model of shape segmentation in which all of these factors (local curvature singularities, global shape properties, and top-down influences) are incorporated. Factors stressed by previous segmentation models thus become integrative parts of a more complex model in which their influences interact with each other.
This PDF is available to Subscribers Only