Purpose - Many computational models of 3D form perception are based on assumptions that can often be violated in natural vision. In the analysis of binocular stereopsis, for example, it is typically assumed that visible surfaces contain identifiable features that can be matched across views. Similarly, in the analysis of shape from shading, it is often assumed that surfaces must be illuminated from above and that they have Lambertian reflectance functions. Our research was designed to examine the importance of these assumptions for human perception. Methods - Observers viewed computer generated stereograms of randomly structured smoothly curved surfaces with specular or Lambertian reflectance functions that were illuminated from above or from the side both with and without identifiable texture elements. Observers estimated local surface orientation at numerous probe points on the depicted objects by adjusting a monocular circular gauge figure until it appeared to rest in the tangent plane at each designated point. Results - For each observer and each condition, we computed a relief map that was most consistent with the local orientation judgments at all of the different probe points. The results revealed a strong linear correlation between the reconstructed surfaces and the actual depicted objects, with an average r2 of approximately .86, though there were large differences in the magnitude of depth scaling in the different conditions. Equivalent amounts of relief were obtained for surfaces that were textured or smoothly shaded with specular highlights, but smoothly shaded Lambertian surfaces appeared significantly flatter. The direction of illumination had no detectable effect on the observers judgments. Conclusions - These findings demonstrate that the perception of 3D shape from shading cannot be based on an assumed overhead direction of illumination or an assumed Lambertian reflectance function, and that compelling stereopsis can be achieved from deformations of shading even when there are no identifiable features to match in each eye's view.
|Journal||Investigative Ophthalmology and Visual Science|
|Publication status||Published - 15 Feb 1996|