Map-image matching using a multi-layer perceptron: the case of the road network

Abstract To help automatize map revision at a scale of 1 : 50,000, a map-guided method is described to update the road network of a map database. This paper describes the essential first step of the procedure, which consists of matching the roads present on both the image and the map database. This matching has to be performed precisely in order to generate meaningful hypotheses on the location of new roads. The matching is conducted by using a multi-layer perceptron (MLP) trained to recognize road segments on the SPOT-HRV panchromatic image corresponding to the cartographic database being treated. Two template matching methods using the trained MLP weight matrix are developed. The first method locates all the road intersections on the image, while the second method locates the segments only. Both methods are not accurate enough to be used alone. However, combining both approaches gives results that are reliable enough to be used in the generation of the hypotheses needed to extract new roads.