Towards Automatic Discovery of Object Categories

Abstract

We propose a method to learn heterogeneous models of object classes for visual recognition. The training images contain a preponderance of clutter and learning is unsuper-vised. Our models represent objects as probabilistic con-stellations of rigid parts (features). The variability within a class is represented by a joint probability density func-tion on the shape of the constellation and the appearance of the parts. Our method automatically identifies distinc-tive features in the training set. The set of model parame-ters is then learned using expectation maximization (see the companion paper [11] for details). When trained on differ-ent, unlabeled and unsegmented views of a class of objects, each component of the mixture model can adapt to repre-sent a subset of the views. Similarly, different component