Korean and English character recognition system using hierarchical classification neural network

Proposes an automatic multilingual character recognition system based on hierarchical classification neural networks for printed Korean (Hangul) and English characters of multi-font and multi-size. The recognition system is designed by hierarchical nets composed of one type classification net and seven type-specific alphabet classification nets. The Korean alphabet is a set of phonetic symbols or alphabets called Jasos that are combined to form a character. The authors introduce the Korean alphabet region segmentation techniques based on the local area blob coloring method. Also the authors propose some reliability factor decision rules in the type-classification net and alphabet classification nets to recover some classification errors. The testing results for 3000 Korean characters and 456 non-Korean characters show that the proposed system generated good solution for multilingual character recognition with about 97% accuracy.