A stochastic model of human visual attention with a dynamic Bayesian network