Evidence against a specialized mechanism for transparent surface perception

Purpose. We identify transparent surfaces easily in natural scenes. Does that mean the human visual system is sensitive to the physical invariants that are specific for transparent surfaces, such as the luminance cues in the figure? Methods. Eight subjects participated in 3-alternative forced choice experiments. They reported whether either the left or right square was perceived as a surface in front of the other, or if neither was seen in front. α_R,β_R,α_L,β_L are the contrasts of the edge sections of the two squares. The stimuli used satisfied α_Lβ_L>0 and α_Rβ_R>0 (Metelli's first rule). Results. 1) Transparent surface segmentation depended on contrast not absolute luminance. 2) A large contrast step along a contour, e.g. |α_R|≫|β_R| indicates that the right square is occluded by a surface (left square) in front. Subjects were sensitive to this cue. 3) When the contrast step was not large, e.g. |α_R| ≈ |β_R|, subjects were not sensitive to sign of the contrast step, i.e. whether |α_R|>|β_R| or |α_R|< β_R. This caused them to perceive a surface that was not consistent with physics, e.g. perceiving the left square as a surface in front when |α_R, < |β_R|. Conclusion. Our results suggest that the visual system may not have a specialized mechanism for the perception of transparency based on physical cues; the most sensitive cue (large contrast step along a contour) is likely to be detected by a mechanism that responds to T-junctions. Transparency stimuli can be used to study general mechanisms of surface segmentation. (Figure Presented).