The Recognition of Printed Korean Characters by ART-Based Neural Network Hierarchy

In the studies of Korean character recognition, the classification of characters by their structural types has been the most common approach; Korean characters have 6 different structural types and are written with 24 letters of the Korean alphabet in rectangular shapes. Because of this structural characteristic of Korean characters, most conventional approaches first classify the structural types of the Korean characters. When the structural types are classified, the component letters are separated. However. with various fonts and sizes, the component letters are often closely connected; and so they are hard to separate and to classify. In this paper, we propose a new approach which is not based on the conventional method of separation of component letters from structural types to recognize printed Korean characters. The new approach is based or! a hier~rchical neural network system consisting of ART(Adaptive Resonance Theory) neural networks and uses diverse features of Korean characters which can be classified irrespective of various fonts and sizes.