The Prax Approach to Learning a Large Number of Texture Concepts

J. Bala, R. Michalski, and J. Wnek, George Mason University

This paper describes an approach, called PRAX, to learning descriptions of a large number of texture concepts from texture samples. The learning process consists of two phases: 1) learning descriptions of selected subset of texture classes, called principal axes (briefly, praxes), and 2) learning descriptions of other classes (non-prax classes), by relating them to the praxes. Descriptions of non-prax classes are expressed in terms of the similarities to praxes, and thus the second phase represents a form of analogical learning. While the first phase is done as a one-step learning process, the second phase is performed as an incremental learning process. The method was applied to learning texture concepts from texture samples, and illustrated by an experiment on learning 24 texture classes, using a subset of 8 classes to learn praxes. After acquiring all texture descriptions from samples taken from a training area, the implemented program, PRAX-2, recognized texture samples from the testing area without a single error.


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.