Identity Management On Homogeneous spaces

We consider the identity management problem, where the identities are classified into two classes, red and blue. The purpose here is to make predictions of the two class identities when confusions arise among identities. In this work, we propose a principle to maintain probability distributions over homogeneous space which provides a mechanism valid for taking into account of any desired degree of approximation. Markov models are used to formulate the two class identity management problem which tries to compactly summarize distributions on homogeneous spaces. Projecting down and lifting up information on different order of statistics can be achieved by using Radon transformations. The commutative property of Markov updating with Radon transform enable us to maintain exact information over different order of statistics. Thus, accurate classification predictions can be made based on the low order statistics we maintained. We evaluate the performance of our algorithms on a real camera network data and show effectiveness of our scheme.