Abstract
Color helps identify objects, but the role it plays in identification relative to other object properties, such as material, texture or shape, is not well understood. To study this, we developed a novel selection-based method. Here we describe how we used it to quantify the relative contribution of color and material to object identification. The stimuli consisted of a set of blob-shaped objects rendered using RenderToolbox3. The objects varied in color (7 levels, from green to blue) and material (7 levels, from specular to matte). One of the objects (level 4 in both color and material) served as the target. On each trial, subjects (N=5) viewed the target together with a pair of test objects drawn from the set. One test object differed from the target in material only (material lure), while the other differed from the target in color only (color lure). Across trials, the size of the material and color differences was varied in a crossed design. Subjects judged which lure appeared more similar to the target and we measured the fraction of trials on which the material lure was chosen over the color lure for each of the material/color difference pairs. We found that subjects exhibit a smooth trade-off between color and material as cues to object identity: the more the color lure differed from the target, the more likely subjects were to choose the material lure. Similarly, the more the material lure differed from the target, the more likely subjects were to choose the color lure. We used the data to quantify the relative importance of color and material cue for each subject. This varied systematically across subjects. Our method enables us to characterize how different object properties are used in identification, and thus measure any reweighting of these properties as their reliability varies.
Meeting abstract presented at VSS 2016