Multi-sensor image registration and fusion

In this dissertation the problems of multi-sensor image registration and fusion are addressed. The research consists of following three parts: (1) Contour-based algorithms for two-dimensional multi-sensor image registration; (2) Registration of three-dimensional tomographic brain scans by curve matching, and (3) Multi-sensor image fusion using the wavelet transform.; Two contour-based methods which use region boundaries and other strong edges as matching primitives are presented for 2-D image matching. The first contour matching algorithm involves the use of chain-code correlation and other shape similarity criteria such as invariant moments. Closed contours and the salient segments along the open contours are matched separately. For the registration of the optical images with noisy Synthetic Aperture Radar (SAR) images, an elastic contour matching approach based on the active contour model is proposed. Experimental results with various kinds of image data have verified the robustness of our algorithms, which have outperformed manual registration in terms of Root Mean Square Error at the control points.; An efficient algorithm for the registration of brain scans from multiple sensory modalities such as magnetic resonance imaging (MRI) and positron emission tomography (PET) is proposed. The method is based on the matching of feature curves which are defined by the intersections of the interhemispherical fissure plane and the skull surface. The matching problem is reduced to a 2-D curve matching task and is solved by a 1-D search scheme.; A multiresolutional image representation associated with the wavelet transform is adopted for image fusion. The wavelet transforms of the input images are appropriately combined, then the new image is obtained by taking the inverse wavelet transform of the fused wavelet coefficients. An area-based maximum selection rule and a consistency verification step are used for feature selection. A performance measure for various image fusion algorithms using specially generated test images is also suggested.