Integration of Assumptions and Cues for Shape from Shading

Abstract

We can perceive shape from images with shading that is a two-dimensional luminance distribution. However it is physically impossible to reconstruct a unique shape from an image with shading. To resolve this ambiguity for perceiving shape from shading, it has been claimed that we use some assumptions, such as "light from above assumption", "single light source constraint" and so on, and cues such as cast-shadow. These assumptions and cues could be inconsistent with each other and the outcome of such a situation is not known. In this research, we investigated how these assumptions and cues work and are integrated for perceiving shape from shading.