Perception of Local Surface Patches Using Shape From Shading

In 1980, Koenderink and van Doorn described a novel idea for representing surfaces, which is based on an inventory of qualitatively distinct patches. These shape categories include dimples, bells, furrows, humps and saddles. The goal of this study was to test whether observers can categorize smooth surface patches based on their parametrically defined categorical features under a wide range of conditions. Method: Parametric surfaces were generated using Gaussian distribution primitives that were projected onto ellipsoids, creating smooth, continuous local patches representing furrows, dimples, bells, humps, and saddle variations. Images of the 3D objects were rendered under a wide range of illuminations, material properties, and scene contexts. For each presented image, subjects performed a categorization task, in which they had to label the depicted patch by selecting one of the possible responses. Results: Subjects performed well under most conditions. However a number of degenerate cases caused confusions between categories, especially between concavities and convexities under restrictive viewing conditions. Conclusion: Our results indicate that observers are reliant on global information to estimate the curvature of local surface patches. Meeting abstract presented at VSS 2014