Alphabetical handwritten address recognition method allowing for no postal code and omission of several address elements

摘要

It is known that a lexicon-driven word recognizer with the database reflected an address class structure is useful to recognize alphabetic handwritten addresses. And, practical performance is realized by narrowing the address class to some level, such as city level, with a postal code recognition result. However, in case of no postal code written and omissions of some address elements, the word recognizer using the lexicon has no correct entries. It causes an increase of address recognition error rates which is an unfavorable factor for practical use. Against these problems, we developed means to control the word recognizer to use a lexicon with a correct entry, and to reject correctly in the case that the lexicon does not include any correct entry. When we applied the means to approximately 2000 actual address images that had the above problems, we could achieve 50% of practical correct sort rate with 0.7% errneous sort rate.