Purpose of this paper is to extend the differential stereo vision algorithm so that it can be used even if left and right intensity levels differ from each other because of camera sensitivity mismatch and/or non-Lambertian surface reflectance. From a multiplicative model of intensity difference, a differential grayness relation including interocular positional disparity (IPD) and interocular intensity disparity (IID) is derived. Applying local least-squares methods to this identity, we obtained a set of efficient local operations to compute the maps of the IPD and IID. From these results, an equation which relates the IID map to a surface reflectance function are derived under assumptions of its spatial uniformity and a parallel incident light. The IID map as well as good satisfaction of this equation are emphasized to be reliable source of binocular luster perception by computer. Experiments for performance tests and applications of this method to image feature extraction and reflectance function identification are shown.