Abstract
In this technical report we make a brief review of the different processes involved in stereoscopic vision (calibration, rectification, correspondence and reconstruction), making a major effort in the correspondence problem. This problem is faced from two viewpoints. First, a hierarchical stereo vision system is proposed where the first step is based in region matching. An algorithm to deal with errors in region-based stereo vision is proposed, which consists of a preprocessing stage, prior to the matching process, where groups of regions in one image that should be matched to the same region in the other image are merged in order to create appropiate regions for the matching. The requirements for merging are analyzed in order to obtain an appropiate similarity criterion in each step of the preprocessing. In order to study the preservation of depth discontinuities, the minimization and regularization approach to stereoscopic vision is also tested and the experimental results with real images are shown.

Keywords:
Stereo, Region-based matching, Oversegmentation, Undersegmentation, Maximal clique, Association graph, Relaxation