Two approaches are being integrated, one is based on regularization-like minimization with a family of different ``stabilizing functionals'' each with an a priori preference for a different type of structure, the other is a more conventional interpolation-based scheme.
The objectives of this research are to: describe objects based on both local and non-local properties, to define algorithms for matching and hence recognizing objects described by three-dimensional surface data, and to elaborate techniques for combining multiple sources of information within this representation framework.
Click here for a gzipped postscript copy of a recent paper on this work.