Recogniton of Hand-Written Character Strings of Mail Addresses by Fusing Two-Stage Character Recognition and Post-Processing

Abstract

In this paper, we first recognize individual character in the prefecture/city part of mail addresses, followed by inferring the entire prefecture/city parts. Then we extend the character category set and recognize entire mail addresses. As pre-processing steps, correction of the characters' stroke widths and orientations is also carried out. This method attempts to improve the recognition rate for individual character as well as the recognition rate for mail address character strings. By using the proposed method, a rate of 85.4% of correctly recognized addresses was achieved in experiments utilizing data distributed by the Research Institute of the Ministry of Postal Affairs.