Abstract
Human lightness (albedo) perception is not necessarily accurate. Previous studies have shown that humans perceive object lightness based on the surface area excluding specular highlights, though this highlight removal performance varies depending on the surface reflective properties and shapes. The purpose of this study was to examine the image features that determine the ability to remove highlights for lightness perception by focusing on the effects of spatial frequency of surface undulations in psychophysical experiments. The stimuli were computer-graphics images of plate-like objects, in which the undulations of real plastic samples were reproduced. There were many conditions in the spatial frequency and amplitude of the undulations, the surface roughness, and illumination maps as experimental parameters. In each trial, a plane-like object (test stimulus) and a matte sphere (reference stimulus) were presented side-by-side on a display. The observer adjusted the lightness of the reference stimulus so that the perceived lightness was matched between them. The results showed that, roughly speaking, the perceived lightness was determined mainly based on the mean luminance of the test stimulus. However, the perceived lightness was lower than that expected from the mean luminance prediction especially on low roughness stimuli, suggesting the contributions of highlight removal. Thus, we defined the difference in the perceived lightness on the test stimulus from that on flat planes as the highlight-removal index. This index moderately correlated with the image luminance contrast from the stimulus images, especially when filtered by the contrast sensitivity function (CSF). However, when the CSF-filtered luminance contrast was smaller than a certain level, the highlight-removal index was fixed at zero even though the undulations on all test stimuli could be easily perceived. These results suggest that perceptual highlight removal for lightness perception requires much more detailed luminance patterns than those for shape perception.