The contribution of local contour features to global shape processing I Fründ, J H Elder |
|---|
Center for Vision Research, York University, Toronto, ON, Canada
|
| How do local properties of bounding contours contribute to visual processing of shapes? We address this question using maximum entropy probability models on the space of simple (non-intersecting), closed contours. We approximated 391 animal shapes by equilateral polygons and analyzed the statistics of the polygons' turning angles. We tested four different models: an unconstrained model, that samples the space of simple, closed contours uniformly, and models that successively constrained the expected angular variance of a shape, the angular kurtosis and the circular correlation between neighboring angles to be equal to the animal shapes. Observers were to select the "more likely natural" one of pairs of shape fragments that differed with respect to one constrained feature. Performance was close to ideal when the fragment pairs differed in circular variance. Performance was at chance when the informative feature was circular kurtosis or circular correlation between neighboring angles. This suggests that observers judge naturalness of shape fragments only partly based on local contour information. We believe that this information is complemented by more global shape features. |
| Up Home |
|---|