Global Image Matching Based on Constraint Satisfaction Neural Network

The key technique to automatically extract the digital terrain model (DEM) from image pairs or stereo pairs is the image matching process. In this paper, the authors describes an approach to using constraint satisfaction neural network to solve the global image matching. The authors firstly give a simple description of the image matching and the constraint satisfaction problem. Then the authors outline an analogy method between image matching process and constraint satisfaction problems and a technique to construct the constraint satisfaction neural network in order to solve the global image matching. The author's utlimate goal is to get the accuracy and robust matching results, given a complicated terrain's aerial photo stereo pairs. So the authors improve the traditional algorithm by use of the new relaxation algorithm presented by Levy 1998. This algorithm can cope with the so called "zero matching" and "multi matching" problem, locate the regions of "zero matching", and bridge the non texture areas. At last, the authors also give some experimental results to show the algorithm's efficiency, accuracy and robust.