Recognition of Handwritten Korean Address Strings by Tight-Coupling of Minimum Distance Classification and Dictionary-Based Post-Processing

In this paper, we propose a system for recognizing handwritten Korean ad- dresses. The input address is composed of four field images, and each field is first segmented into individual characters and then recognized by a character classifier in cooperation with a dictionary-based post-processing. In contrast to conventional approaches to word recogni- tion, we have combined the character classifier and post-processor as tightly-coupled as pos- sible to compensate the weakness of each other and to get a high recognition performance in the address interpretation. Higher performance of the proposed system has been proven by an experiment with a database of handwritten Korean addresses, in which more than 80% of 1,107 addresses are processed with less than 1% of reading error. Its processing speed is about two times faster than those of conventional approaches without post-processing. The proposed system can be useful for the interpretation of some constrained handwritten Korean addresses in forms, and can be extended easily to the recognition of unconstrained ones in general forms and mail envelops.