Since the conventional feedforward neural networks for character recognitio
n have been deigned to classify a large number of classes with one large ne
twork structure, inevitably it poses the very complex problem of determinin
g the optimal decision boundaries for all the classes involved in a high-di
mensional feature space. Limitations also exist in several aspects of the t
raining and recognition processes. This paper introduces the class modulari
ty concept to the feedforward neural network classifier to overcome such li
mitations. In the class-modular concept, the original K-classification prob
lem is decomposed into K 2-classification subproblems. A modular architectu
re is adopted which consists of K subnetworks. each responsible for discrim
inating a class from the other K-1 classes. The primary purpose of this pap
er is to prove the effectiveness of class-modular neural networks in terms
of their convergence and recognition power. Several cases have been studied
, including the recognition of handwritten numerals (10 classes), English c
apital letters (26 classes). touching numeral pairs (100 classes), and Kore
an characters in postal addresses (352 classes). The test results confirmed
the superiority of the class-modular neural network and the interesting as
pects on further investigations of the class modularity paradigm. (C) 2001
Pattern Recognition Society. Published by Elsevier Science Ltd. All rights
reserved.