Abstract

When people have to extract parts from a new category of objects, where they do not know the shape of the parts, they must extract shape from the raw data. If the parts composing the objects transform from one instance to another, part extraction is difficult because any part could potentially correspond to any other part in the object category. However, our perceptual apparatus rules out many correspondences. We investigated the role of sign of curvature in part extraction in categories of 2D B-spline stimuli. Subjects were exposed to one category of objects composed of a random and a target part whose shape was unknown a priori. In the concave and convex conditions, the target part was transformed, but its signs of curvature remained constant throughout the category, while in the mixed condition, the transformation included a change of sign. Results showed that subjects in the concave and convex groups extracted a single part from the category while subjects in the mixed group extracted two parts: one with a positive and one with a negative sign of curvature.

© 1992 Optical Society of America