|| Checking for direct PDF access through Ovid
Humans are able to estimate the reflective properties of the surface (albedo) of an object despite the large variability in the reflected light due to shading, illumination and specular reflection. Here we first used a physically based rendering simulation to study how different statistics (i.e. percentiles) based on the luminance distributions of matte and glossy objects predict the overall surface albedo. We found that the brightest parts of matte surfaces are good predictors of the surface albedo. As expected, the brightest parts led to poor performance in glossy surfaces. We then asked human observers to sort four (2 matte and 2 glossy) objects in a virtual scene in terms of their albedo. The brightest parts of matte surfaces highly correlated with human judgments, whereas in glossy surfaces, the highest correlation was achieved by percentiles within the darker half of the objects’ luminance distributions. Furthermore, glossy surfaces tend to appear darker than matte ones, and observers are less precise in judging their lightness. We then manipulated different bands of the virtual objects’ luminance distributions separately for glossy and matte surfaces. Modulating the brightest parts of the luminance distributions of the glossy surfaces had a limited impact on lightness perception, whereas it clearly influenced the perceived lightness of the matte objects. Our results demonstrate that human observers effectively ignore specular reflections while evaluating the lightness of glossy objects, which results in a bias to perceive glossy objects as darker.