Abstract
Light reflection of glossy dielectric materials consists of specular and diffuse components. The spectral distribution of illumination is preserved in the specular reflection, while modulated by surface light absorption for the diffuse reflection. This leads to the following physical constraint: the specular spectral distribution should include the diffuse spectral distribution, and highlights (specular + diffuse) should be brighter than the surrounds (diffuse only) when seen through any narrow band filter. By changing the color combination of specular and diffuse components, we have shown that human observers perceive naturalistic glossy surfaces when the physical constraint holds (e.g., red highlights on red body, white on red), but not otherwise (e.g., red on white, red on green) (Nishida et al., VSS2008). We further observed that even if the color combination is valid (e.g., white on red), the gloss perception is lost when highlights have no luminance increments. A hypothesis of gloss computation consistent with the physical constraint and the observed perceptual effects is to decompose the image into multiple color bands, and run achromatic gloss analysis at each band. We tested this idea with using L-cone, M-cone and S-cone images as biologically plausible band images. We made a display in which the L- and M-cone images contained a glossy object with natural bright highlights, while the S-cone image the same object with unnatural dark highlights. The resulting image violated the physical constraint, but the display looked naturally glossy when the S-cone image intensity was adjusted so as to render highlights apparently white. The S-come image was not simply discarded, since the display looked unnatural with different S-cone image intensities. Multiple color band analysis is a promising hypothesis of luminance-color interactions, but our observation indicates that the simplest version using raw cone images cannot fully explain the luminance-color interaction in gloss perception.
Grant-in-Aid for Scientific Research on Innovative Areas.