| Several topics in the area of image, or more generally, data registration are investigated. An important problem in image registration is to find the best spatial fit between two or more images that overlap the same scene. One of our primary goals is to be able to register images which arise from different modalities. Geometric models are described which govern how images may spatially differ. Corresponding to each model, a method to determine the parameters of the model is presented. Each of these methods is based on using collections of computationally generated ground control points of the images being registered. Among the transformations addressed are various affine transformation models, including rotation, translation, and scale. Two models of geometric distortion which arise from varying aspect angles are covered in detail. These models simulate the distortion caused by non-vertical views of a scene with either an orthographic or a perspective view. Using methods to determine the corresponding parameters of the model, hot pixel methods are then formulated to approximate optimal parameters via a computationally efficient approach. |
