Automated recognition of handwritten addresses on mail-pieces

This paper describes a strategy for the automation of reading handwritten addresses which is one task of postal OCR-readers. Main emphasis is put on handwritten addresses with connected cursively written words, also called script words. In comparison with the analysis of machine printed addresses a completely different and much more flexible approach has been developed making intensive use of all available context knowledge. Special script recognition methods have been developed to process a word as a whole in addition to those methods which handle single characters. The system requirements and its solutions are outlined by a variety of examples.<>