Three-dimensional vision measuring technique aims at automatically reconstructing the properties of object and scene such as shape, motion, and deformation from one or more images or video clips in the real world around us, so that three-dimensional true environment can be recognized and understood. To solve the problems such as brightness change, scale change, rotation change and noise in conventional approaches, a new three-dimensional vision measuring technique is proposed. Firstly, the multi-modal isotropic monogenic features which include geometric feature (local orientation), structure feature (local amplitude and local phase), color phase and color texture are extracted by Poisson kernel and Riesz transforms. Secondly, a novel multi-modal feature descriptor and a similarity measurement function are constructed by simulating human vision function. Finally, the proposed similarity measurement function is applied to stereo matching and three-dimensional vision reconstruction. At the same time, the qualitative and quantitative evaluations are finished on the synthetic and natural stereo images.