Abstract
The visual system can easily distinguish between a "mirror" object, which has a perfect specular surface like a polished metal, and a "glass" object, which has a transparent and refraction medium like ice under natural illumination, although an observer just observes a complex distorted image on the object surfaces from the surrounding illumination. In order to identify the visual cues to distinguish mirror and glass materials, we investigated a relationship between observers' performances and static/dynamic cues in visual stimulus.. Three randomly-bumped (potato-shape) objects were modeled by Blender 2.77 and were rendered as mirror and glass material objects by Mitsuba renderer (Jacob, 2010) under five real-world illuminations (Debevec, 2000; Adams et al., 2016). To examine both static and dynamic effects, we presented the static objects as image or the horizontally rotating objects as a video. The observer's task was to identify its material (mirror or glass) in a 2AFC paradigm. In this result, we found that a luminance distribution along a vertical direction in the image was significantly correlated with the performance in the static condition, and a ratio between positive and negative values of motion components along the object rotating direction in the video was significantly correlated with the performance in the dynamic condition. Moreover, for the stimuli color, which inverted from positive to negative values, the observers relatively more depended on a dynamic cue. This result can be caused by a decrease of a reliability of the static cue by the color inversion. These results suggest that the visual system effectively and comprehensively uses multiple cues in order to distinguish several materials in various contexts.
Meeting abstract presented at VSS 2017