Recognition of Character Strings Printed with Large Alignment Error

Optical character reader (OCR) technology for reading documents, such as monetary transaction documents, is becoming more and more important than ever before.The position of the printed character string is sometimes largely shifted from its designated position, and there may be two or more directions in spite of one sheet of paper. There are various causes, the performance of the printer, a mistake in the printing position designed by the software, a variation in the cell positions caused by the publishers and by the publishing dates, or even a mistake of the handwriting position. We developed a recognition method for determining which character strings are to be read in such difficult situations. A method for determining the correspondence of character strings to cells with a very high success ratio was developed. This method is based on local and global rules, and effective control of these rules. It was estimated to have a 99.2% success ratio using an experiment on 11,387 character strings.