Learning Structural Models in Multiple Projection Spaces

In this paper, we present an algorithm for learning structures of Bayesian models in multiple projection spaces. We assume that a visual phenomenon can be projected on a set of spaces that share a common subspace. We propose that models of individual projections can be related through probability distributions over the shared subspace. We develop a learning method that estimates simultaneously the structure and parameters of an integrated model of the target phenomena. This integrated model combines information from all individual projections. The model learning procedure is accomplished by maximizing the Bayesian Information Criterion within the setup of the Expectation-Maximization algorithm. Finally, we show how the method can be applied to the problem of learning and recognizing human motions.