The Perception of Shading and Reflectance

Edward H. Adelson and Alex P. Pentland

Published in
Perception as Bayesian Inference (pp. 409-423)
Eds: D. Knill and W. Richards
New York: Cambridge University Press (1996).
Traditional models of lightness perception have been designed to deal with scenes consisting of patches on a flat plane. Yet the perception of lightness is highly dependent on the perceived three dimensional configuration of objects. A successful model should be able to decompose an image into a set of intrinsic images which capture scene characteristics such as reflectance and illumination. We describe an approach for dealing with some simple polyhedral scenes. A set of "specialist" processes are used, each of which captures information about certain aspects of image formation. The specialists interact in such a way as to find a description of the scene that minimizes a cost function, where the cost function is related to the simplicity or likelihood of a given interpretation. We use this approach to analyze some simple scenes in terms of reflectance, 3-D shape, and lighting direction.