Recognition of Hand-Printed Chinese Characters and the Japanese Cursive Syllabary

We describe a method for the recognition of hand-printed Chinese characters (Kanji) and Japanese cursive syllabary characters (Hiragana). The system preclassifies using cellular features and then classifies by a combination of feature extraction of line segments, an extreme-point method, and a relaxation matching method. Relaxation is a technique using contextual information to reduce local ambiguities: initial probabilities are assigned to matches between pairs of the line segments in the dictionary and the input, and then iteration finds acceptable combinations of the matches. The system was trained and tested on the ETL8 database; recognition results were 99% correct on the “good quality” data set, and 94.6% on the “poor quality” data set.