The motivation of this study is to explore the ability of human visual phase discrimination for stereo vision. A very important issue is to understand how much information is carried by the phase information in human stereo interpretation

hence helping for creating new models and approaches which are efficient for interpreting natural scence. Based on the visual cue

we have designed a new approach of stereo vision utilized the wavelet theory. Unlike the area-based that is computationally expensive

and also the feature-based which generates only sparse disparity map

the approach uses the image phase as the matching primitives which can generate a dense disparity field efficiently. In order to deal with complicated natural scenes

the constraints controlling the surface discontinuity and occlusion are also introduced in our approach. Experiments have proved that the proposed approach can be adaptable for almost arbitrary images and resulting disparity surface is accurate and dense. From this study we could probably understand what role of phase information is in human stereo fusion and help developing appropriate computational approach for practices.