Recognition of Cursive Handwritten Country Names on Overseas-Addressed Mail

Background: Mail items posted within a country are either addressed for internal (within-country) delivery or to overseas destinations. Overseas-destined mail must be pre-sorted in the country of origin and forwarded to the destination country for further sorting and delivery. Handwritten addresses, which form a considerable percentage of all postal mail, pose a particularly difficult problem for automatic sorting. This paper presents an approach to automatic sorting of overseas destined mail items by locating and recognizing the handwritten country name. Method: The method proposed extracts structural features from the handwritten country names and combines this with OCR of the first letter of each word to solve the problem of limited vocabulary off-line unconstrained handwritten word recognition. The emphasis is to create a simple and reliable method of recognition, which can be readily implemented, in a fast real-time sorting system. Preprocessing, word holistic feature extraction, character classifier, and integration of lexical and syntactical knowledge are used in this system. Results: An accuracy of 85.0% correct country name recognition was achieved with an error rate (wrongly recognized country name) of 0.5% and a rejection rate (could not recognize the country name) of 14.5% using a test set of 1294 address images of the 371 variations of country names observed on overseas-addressed mail. Conclusions: The proposed method demonstrates the feasibility of fast and accurate country name location and sorting.