Verification techniques for high-performance OCR of hand-printed postcodes

To minimise error-rate in the recognition of handwritten addresses, redundancy between the postcode and the remainder of the address must be exploited. The paper investigates methods by which postcodes can be verified using features extracted from the remainder of the address. Features which define the town name effectively, and can easily be extracted from a handwritten or hand-printed town name are proposed. Experiments to determine the performance obtainable in extracting these features using a database of 820 constrained address images are reported. An overall verification rate of at least 47% for handprinted addresses and 38% for handwritten addresses is predicted, based upon cascading the feature extraction processes described