Search

Human observers estimate the illumination direction of rough surfaces rather precisely. When surfaces are rough, the illumination generates visible “texture” from differential shading at the level of the roughness, whereas differential illumination at the level of significant global surface curvature leads to the more familiar “shading.” The shading is used in conventional shape-from-shading (SFS) algorithms, which ignore the illumination texture cue. Because of this simplification, SFS algorithms are typically formulated as global problems (partial differential equations, etc.). Human observers are likely to apply different methods than do these conventional SFS algorithms, however. When the roughness is not isotropic, one expects systematic errors in the visual detection of illumination direction, conceivably giving rise to erroneous shape estimates. Here we addressed this issue through systematic psychophysics on illumination direction detection as a function of the roughness anisotropy. Our expectations were fully borne out, in that the observers committed the predicted systematic errors. These results are precise enough to allow the inference that illumination direction detection is based on second-order statistics—that is, of edge detector (rather than line detector) activity. [ABSTRACT FROM AUTHOR]