A System for Mailpiece ZIP Code Assignment through Contextual Analysis. Phase 2

Abstract : This report describes the continued development and testing of a system for contextual analysis of machine printed address block images. The system receives a binary image of the address block (location of the address block is not a part of this work) and then: (1) segments the image into lines, words, and characters, (2) parses the address block to assign roles to the various words, (3) Forms queries based on number of words in the street name, word lengths, identification of suffix and directional words, and (externally supplied) knowledge the ZIP code on the mailpiece, (4) retrieves street records from a postal directory, (5) matches the street names from the directory to the word images using a character confidence measure, (6) given a close match uses (externally supplied) knowledge of the street number to assign a 4 digit add-on code. It should be stressed that this system does not use isolated character recognition to read words, but rather a lexicon based word verification system. The report emphasizes description of the segmentation processes, an automatic feature generation method used in the word verification system, and the structure of the postal database.