Empirical design of a holistic verifier for automatic sorting of handwritten Singapore postal addresses
Chin Keong Lee, Graham Leedham · 1999
This paper describes an algorithmic architecture of an optical character recognition (OCR) system integrated with a Singapore handwritten address interpretation (SHWAI) system. The proposed OCR-SHWAI system generates multiple hypotheses of postcodes, verifies the hypotheses using a postal dictionary, and uses address features to choose among the hypotheses. The performance of this system on a set of 450 fictitious but realistic handwritten mail pieces shows an improvement in sorting performance from 81.6% correct, 6.0% reject, 12.4% error using OCR alone on the postcode to 70.9% correct; 28.5% reject; 0.7% error representing a significant improvement in the error rate.