Robust image registration for fusion

Fusion can be seen as the process of forming a uniform description of multiple information sources. The type of description depends on the application domain as well as the type of features to be considered. Features (also known as “descriptors”) relate to iconic, symbolic, semantic or pragmatic aspects of some physical phenomenology of the real world. The process of feature extraction results in a feature map and can be top-down or bottom-up or a hybrid mixture of both. In the next step the correspondence problem (matching) of multiple feature maps has to be solved. This paper concentrates on symbolic feature extraction of straight line segments forming the feature map and their robust and precise matching with other feature maps. Together with extensions concerning point-like features these approaches form the basis of many real-time capable applications. Three of them will be described: image-to-image registration, automatic geocoding, and change detection for industrial inspection and quality assurance.