Transparent objects: Influence of shape and color on depth perception

Recovering depth information from a single still image is an important problem in computer vision. However, the problem is difficult and challenging because it has an infinite number of solutions. To address this issue, humans use numerous visual cues to infer depth. Much progress has been made towards an understanding of the visual mechanisms involved in 3D perception. Such an understanding provides relevant knowledge to design efficient approaches for computer vision. While there is much prior work on opaque objects, there has been relatively little in relation with transparency. In this paper we investigate depth estimation from single still images containing nonplanar and real transparent objects. We focused our study on two visual features: shape and color. A database of stimuli was created to carry out a psychophysical experiment with 42 naïve observers.