A pair of stereo images can be viewed to perceive a 3D scene. For the human vision system this is an easy process for both real images and random dot stereograms. An equivalent 3D machine vision system using local correlation can be used for decoding of random dot stereograms and for producing edge images of stereo image pairs. If the objects in the stereo images have sharp edges, the depth of these can be revealed by local correlation and threshold. The depth contour of the connecting surface is more difficult to detect and is, with the current technique, absent or disrupted. In the present paper the local correlations technique and the estimation of the depth are discussed and results from real stereo images are shown.