Probabilistic models of appearance for 3-D object recognition

We describe how to model the appearance of a 3-D object using multiple view s, learn such a model from training images, and use the model for object re cognition. The model uses probability distributions to describe the range o f possible variation in the object's appearance. These distributions are or ganized on two levels. Large variations are handled by partitioning trainin g images into clusters corresponding to distinctly different views of the o bject. Within each cluster, smaller variations are represented by distribut ions characterizing uncertainty in the presence, position, and measurements of various discrete features of appearance. Many types of features are use d, ranging in abstraction from edge segments to perceptual groupings and re gions. A matching procedure uses the feature uncertainty information to gui de the search for a match between model and image. Hypothesized feature pai rings are used to estimate a viewpoint transformation taking account of fea ture uncertainty. These methods have been implemented in an object recognit ion system, OLIVER. Experiments show that OLIVER is capable of learning to recognize complex objects in cluttered images, while acquiring models that represent those objects using relatively few views.