The recognition of handwritten Chinese characters from paper records

Abstract

This paper describes a method used for the recognition of handwritten simplified Chinese characters from paper records. The method is based on the use of discrete hidden Markov models. The recognition accuracy achieved for all 3755 common simplified Chinese characters in the GB1 character set is 91.2% for top 1 choice and 98.5% for top 5 choice. The method recognizes isolated characters only and not words or phrases. The test set contained about 35,000 characters. All characters were written in a print style.