To the Graduate Council: I am submitting herewith a dissertation ...

AbstractThe focus of this research is on building 3D representations of real world scenes andobjects using different imaging sensors. Primarily range acquisition devices (such as laserscanners and stereo systems) that allow the recovery of 3D geometry, and multi-spectralimage sequences including visual and thermal IR images that provide additional scenecharacteristics. The crucial technical challenge that we addressed is the automatic pointsetsregistration task. In this context our main contribution is the development of anoptimization-based method at the core of which lies a unified criterion that solvessimultaneously for the dense point correspondence and transformation recovery problems.The new criterion has a straightforward expression in terms of the datasets and thealignment parameters and was used primarily for 3D rigid registration of point-sets.However it proved also useful for feature-based multimodal image alignment. Wederived our method from simple Boolean matching principles by approximation andrelaxation. One of the main advantages of the proposed approach, as compared to thewidely used class of Iterative Closest Point (ICP) algorithms, is convexity in theneighborhood of the registration parameters and continuous differentiability, allowing forthe use of standard gradient-based optimization techniques. Physically the criterion isinterpreted in terms of a Gaussian Force Field exerted by one point-set on the other. Suchformulation proved useful for controlling and increasing the region of convergence, andhence allowing for more autonomy in correspondence tasks. Furthermore, the criterioncan be computed with linear complexity using recently developed Fast Gauss Transformnumerical techniques. In addition, we also introduced a new local feature descriptor thatwas derived from visual saliency principles and which enhanced significantly theperformance of the registration algorithm. The resulting technique was subjected to athorough experimental analysis that highlighted its strength and showed its limitations.Our current applications are in the field of 3D modeling for inspection, surveillance, andbiometrics. However, since this matching framework can be applied to any type of data,that can be represented as N-dimensional point-sets, the scope of the method is shown toreach many more pattern analysis applications.