Abstract
To estimate shape from shading, the visual system may use the knowledge about the relationship between image intensity and surface orientation. This knowledge is called the reflectance map, and can be considered as a kind of natural constraint. The physical relationship between image intensity and surface orientation varies depending on the surface reflectance function of the viewed object and the lighting condition. Thus, depending on these factors, the visual system needs to modify the reflectance map to estimate shape from shading accurately. It is already known that the visual system modifies the reflectance map concerning the light source direction. In this study, concerning the surface reflectance function, we investigated whether the visual system adaptively modifies the reflectance map, or it always assumes a general-purpose surface reflectance function such as Lambertian function. We presented subjects various images of cylindrical surfaces with different cross-sectional shapes and shading parameters (Blinn's shading model was used to render the stimuli). However, the light source direction was always fixed at the viewer's direction. Then, the perceived shapes from these images were measured. The reflectance maps used by the visual system for these stimuli were estimated from the relationship between image intensity of the stimuli and slant of the perceived shapes (Seyama & Sato, 1998). As a result, the estimated reflectance maps significantly differed from each other. Since the light source direction was fixed during the experiment, the difference among the estimated reflectance maps suggests that the visual system modified the reflectance map concerning, most likely, the reflectance function. These results indicate an adaptive nature of the shape from shading performed by the visual system.