Statistics in the Image Domain for Mobile Robot Environment Modeling Abstract This paper addresses the problem of estimating dense range maps of indoor locations using only intensity images and sparse partial depth information. Unlike shape-from-shading, we infer the relationship between intensity and range data and use it to produce a complete depth map. We extend prior work by incorporating geometric information from the available range data, specifically, we add surface normal information to reconstruct surfaces whose variations are not captured in the initial range measurements. In addition, the order on which we synthesize range values is based on a best-first approach that uses edge information from the intensity images, and isophotes lines from the available range. Our method uses Markov Random Fields to learn the statistical relationships from the available intensity and from those sparse regions where both range and intensity information is present. In contrast to classical approaches to depth recovery (i.e. stereo, shape from shading), we make only weak assumptions regarding specific surface geometries or surface reflectance functions since we compute the relationship between existing range data and the images we start with. Preliminary results on real data demonstrate that our method works reasonably well when incorporating geometric information.