Object class recognition by unsupervised scale-invariant learning

TLDR

Objects are modeled as flexible constellations of parts. We present a method to learn and recognize object class models from unlabeled, unsegmented cluttered scenes in a scale‑invariant manner. The method employs a probabilistic representation of shape, appearance, occlusion, and scale, selects regions with an entropy‑based detector, estimates model parameters via EM in a maximum‑likelihood framework, and classifies images Bayesianly. The model achieves excellent results on diverse datasets, including geometrically constrained classes such as faces and cars, and flexible objects such as animals.

Abstract

We present a method to learn and recognize object class models from unlabeled and unsegmented cluttered scenes in a scale invariant manner. Objects are modeled as flexible constellations of parts. A probabilistic representation is used for all aspects of the object: shape, appearance, occlusion and relative scale. An entropy-based feature detector is used to select regions and their scale within the image. In learning the parameters of the scale-invariant object model are estimated. This is done using expectation-maximization in a maximum-likelihood setting. In recognition, this model is used in a Bayesian manner to classify images. The flexible nature of the model is demonstrated by excellent results over a range of datasets including geometrically constrained classes (e.g. faces, cars) and flexible objects (such as animals).