This paper investigates verification schemes and their applications to the recognition of both isolated and touching handwritten numerals. Definitions and functionality analyses of the verifiers are given. The measurement of precision rate is used to assess the system reliability in a class-specific manner. Verification-enhanced systems are proposed with extensive experiments conducted on both isolated and touching numerals. Two databases for touching numerals are built/organized to serve as standard data sets. Experimental results indicate a substantial improvement in system precision rates by the verification scheme, which proves the effectiveness of the proposed systems and justifies the important role of verifiers in OCR systems. (C) 2002 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. [References: 23]
展开▼