Edge-and Shape-Based Geometric Registration

The standard method for geometric registration of images consists of selecting control points in the two images and computing the correlation maximum of small subimages containing the control points. This method does not work well when applied to images taken at different seasons or with different sensors. The use of edge-based registration has been proposed to overcome these difficulties but has so far achieved no better than picture raster element accuracy. This paper presents edge-and shape-guided correlation (or comparison) of control point areas for the analysis of multitemporal and multisource data. The direct correlation of control areas for registration is supplemented by comparison of descriptions of elementary objects, e. g., drawn lines, borders, and edges, whose positions are known with subpixel accuracy. These methods have been implemented as a set of image registration modules within the context of the DIBIAS image processing system.