Visual Learning: An Overview

We review modern approaches to the learning and recognition of complex patterns, including discriminant functions, neural networks, decision trees, and hidden Markov models. We then intro duce several relational learning systems and discuss in detail one specific technique, conditional rule generation. This technique is shown to be very flexible and useful for the learning of static patterns, such as objects, as well as dynamic patterns, such as movement patterns. The technique is illustrated with a number of very difficult visual learning problems.