Learning and Upgrading Rules for an OCR System Using Genetic Programming

:Rule-basedsystemsused for OpticalCharacterRecognition (OCR)are notoriously difficultto write,maintain,and upgrade. Thispaperdescribesa method forusing Genetic Programming(GP) to evolveand upgrade rules for an OCR system. The languageof the evolved programs was designedsuch that humanhand-coded rulescanbe included intothe initial populationin orderto upgrade for a new font. Thesystemwas successfulatlearningrules for largecharactersetsconsistingof multiplefontsand sizes,with very good generalizationto test sets. In addition,the methodwas found tobe successfulatupdating hand-coded rules written in C for new fonts. Thisresearch demonstratesthe successfulapplicationof GP to a difficult,noisy, real-world problem.

• Type: Conference
• Year: 1994
• Where: ICEC
• Authors: David Andre
• Keywords: And Upgrade, Difficultto Write, ICEC 1994, Maintain