Abstract
In daily life, we encounter a variety of objects made of metals including gold, silver, and aluminium, and easily distinguish them from non-metallic stuff. For example, we perceive gold as clearly different from glossy yellow, or silver from glossy gray. In physics, the metallic surface is typically defined as having an extremely strong specular reflectance relative to the diffuse reflectance. Here, we examined if and how human judgement of metallicity depends on this parameter under variable illuminations. Using an HDR display (BrightSide DR37P; 0.1 − 3500 cd/m2), we presented a list of seven computer-generated bumpy objects with variable specular/diffuse reflectance ratios simulated under variable scenes and levels of image-based illumination, and asked the observers to classify them into metallic and non-metallic ones. When the illumination level was low, the observers often judged mirrored objects as non-metallic, depending on the scene of illumination; e.g., a mirrored surface illuminated by few ramps in a church looked glossy, but hardly looked metallic. When the illumination level was comparably high with the real world (Lmax [[gt]] 2000 cd/m2), on the other hand, the classification depended on the specular/diffuse ratio regardless of the scene (except a snow field). The judgement data, as well as the physical specularity, for objects under a fixed illumination was characterized by changes in the standard deviation, but not skewness, of the image luminance histogram. For most scenes, the effect of illumination level appeared to be related to the image mean luminance as an anchor. These results suggest a possibility that the perception of metallic surfaces is determined by simple image statistics, but in a different way from the perception of mere glossiness that mainly depends on skewness (Motoyoshi, Nishida & Adelson, 2005, JOV, 5, 569a).