Automated spline surface modeling and matching for recognition of free-form objects

Surface representations are utilized in a multitude of applications such as computer vision, medical imaging and computer graphics. B-spline surfaces have a number of desirable properties for representing surfaces, however, the complexity of knot placement strategies has prevented their widespread use in high-level vision environments. A solution to this problem is formulated within the reversible jump Markov chain Monte Carlo framework, whereby a derived posterior distribution may be sampled to calculate expected values for the number of knots required, their expected positions and a maximum likelihood estimate for the resulting control net of a given surface. Recognition of individual models may then be achieved using a hash table constructed using the principal components of the model control nets. Results of the fitting procedure, in terms of estimated knot vectors and spline surface errors, and the recognition of objects are provided for a set of free-form objects.