STATISTICAL OPTIMIZATION AND GEOMETRIC INFERENCE IN COMPUTER VISION

This paper gives a mathematical formulation to the computer vision tas k of inferring three-dimensional structures of a scene based on image data and geometric constraints. Introducing a statistical model of ima ge noise, I define a geometric model as a manifold determined by the c onstraints and view the problem as model fitting. I then present a gen eral mathematical framework for proving optimality of estimation, deri ving optimal schemes, and selecting appropriate models. Finally, I ill ustrate the theory by applying it to curve fitting and structure from motion.