Hopfield network for stereo matching of panchromatic urban IKONOS images

Stereo matching is the object of many researches, and the various approaches depend primarily on the constraints used and the applications sought such as mobile robotics, medical imagery, virtual reality and remote sensing. It is in the last scope of application which we propose in this article the use of a Hopfield network for the stereo matching of pairs of urban panchromatic satellite images taken from IKONOS2 satellite. We are interested in “build” and the primitive used to match the images is the region, for that, a region segmentation is applied initially to the couples of images then followed by a region matching process. From there, we use a first approach which consists in using a classical matching. The second approach consists in applying a Hopfield neural system initialized by a classical matching, this network has the particularity of using new constraints, compared to the first approach, the neural method provides better results: a high matching rate and a good reduction in ambiguities.