Map-image registration using automatic extraction of features from high resolution satellite images

• Main Author: Vohra, Vijay Kumar
• Published: University College London (University of London) 1999
• Subjects: 538.7; Geomagnetism
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.313051

In every part of the world the rate of map revision is alarmingly low, when compared to the rate of change of many human influenced surface features. Map making is very time-consuming and often information used for updates has become history before the updated map is made available. There is therefore a requirement to regularly gather up-to-date information about surface features and to incorporate changes in maps both quickly and efficiently. Automation of two systems, i.e. the automation of map-image registration and then of change detection can fulfill the requirements of map revision. This thesis works on the first system. The piece of work in this study has looked into a fast and an accurate solution to register high resolution satellite images to maps. This will allow changes in ground features to be used to update maps. Photogrammetric techniques used to update maps have previously shown good results, but they are tedious, time-consuming, and not beneficial for updating small changes at all. Feature extraction methods were used in the present study. The system developed was designed for automatic extraction of suitable areal features in images. The emphasis was on areal features rather than point or linear features because they have a distinctive shape, and they are extracted easily from vector as well as raster data. The extraction of suitable polygons, as control information, from images was obtained by using two matching techniques. Patch matching to extract the conjugate map and image polygons, and dynamic programming to find the corresponding matched boundary pixels of the map and image polygons. Some matched points were incorrect because of perspective, shadows and occlusions. A statistical model was developed to remove perspective distortion and large errors. The model demonstrated the removal of erroneous match points, and selected the good match points and registered the images to maps with a sub-pixel accuracy. A novel aspect of the study is that the automation is achieved with high accuracy in flat and moderate terrain areas without using height information, as it is essentially used in photogrammetric techniques.