Abstract

Illumination tends to vary in natural visual scenes, incorporating well-lit areas and darker regions. These darker areas have lower relative contrast and may therefore provide less detailed information. Recent research in the lightness domain, both using eye-tracking methodology and psychophysics, confirmed that there is a tendency to avoid shadowed regions when estimating color. We investigated a similar tendency but instead employed high level visual judgment tasks. In the two experiments we presented photographs of human faces half covered with a shadow. In the first experiment, we digitally modified photographs in order to obtain well-controlled and equal distributions of lower contrast areas, while in the second experiment we used photographs that already contained natural shadows. We asked our participants to judge beauty, age and profession of the depicted people. Using a state-of the-art eye tracker, we measured the first fixation, the number of fixations and the dwell time, both at the level of the whole image and for pre-defined regions of interest (such as the eyes, the mouth, etc.). For all measures and for all tasks, a clear preference to fixate the non-shadowed regions was found. This suggests that visual sampling preferences for well-lit areas extend beyond low-level visual tasks.