Nonparametric Bayesian attentive video analysis

We address the problem of object-based visual attention from a Bayesian standpoint. We contend with the issue of joint segmentation and saliency computation suitable to provide a sound basis for dealing with higher level information related to objects present in dynamic scene. To this end we propose a framework relying on nonparametric Bayesian techniques, namely variational inference on a mixture of Dirichlet processes.