High performance character recognition for off-line handwritten British postcodes

Provides details of a dictionary search scheme which efficiently improves the overall postcode recognition rate. The research described has shown that it is possible to extract high word recognition rates from an unoptimised character recognition algorithm, by using contextual knowledge. A basic character recognition rate of 63.1% was achieved for the full alphanumeric character set, as written in a database of British postcodes. The addition of a syntactic recognition stage, based upon analysis of allowable British postcode formats, improved the character recognition rate to 73.2%. The effect of syntax was mainly to reduce the confusion between numerals and alphabetic characters. A trie-structured dictionary was used to add contextual knowledge to the postcode recognition process.<
>