Abstract
The presence of gloss on the surface of objects is a challenge for the visual system when trying to recover their color (Xiao and Brainard, 2008). Here we test how humans deal with glossy objects for judging their lightness. Observers were presented with two rendered images of golf balls and had to judge which one had a lighter surface. The balls could be either glossy or matt. One of the two objects had a constant reference reflectance while the reflectance of the other one was varied in nine steps. The slopes of the discrimination curves were shallower (by more than a factor of 2) for the glossy objects, indicating lower performance. We observed previously (Toscani,Valsecchi and Gegenfurtner, VSS 2011) that observers fixated the brightest parts of matte objects when performing a similar task. Using a physically based rendering simulation we demonstrated that for matt objects the diagnostic value of image pixels about the objects' reflectance increased monotonically with their luminance. For glossy objects, however, the brightest pixels are of course not informative of the objects' reflectance. In order to investigate whether observers behaved accordingly, we recorded their eye movements while doing the task. Consistent with the results of the simulation, observers had a tendency not to look at the specular highlights when asked to judge the lightness of glossy objects. In contrast, a bottom up saliency approach (Itti and Koch, 2000) applied to our glossy objects predicted that the highlights should be preferentially fixated. In summary, the data we collected from glossy objects show that observers sample the most informative parts of objects for making lightness judgments. They use an optimal strategy without being distracted by the highly salient but less informative specular highlights.
Meeting abstract presented at VSS 2013