Robust multihypothesis discrimination of controlled i.i.d. processes

This paper describes a general framework for the robust discrimination of objects represented as a family of i.i.d. random distributions. Testing is based on accumulating evidences on the discrimination between all-pairs of hypotheses by sampling the family of distributions according to an optimal control law. The optimality criterion is built on constraint satisfaction issues. An application on 2D rotation invariant shape recognition with noisy contours illustrates the approach.