A System to Locate and Recognize ZIP Codes in Handwritten Addresses

A system to automatically locate and rccognize ZIP Codes in handwritten address is described. Given a grty-scale image of a handwritten address block, the system prcprocesses the image by thrcsholding, border rcmoval and underline rcmovrl. One or morc candidrte words for the ZIP Code arc isolated. Each csndidate is divided into 5 or 9 segments and rccognition ls attempted on each segment. Digit rccognition is accomplished by means of an arbitradon procedurc that takes as input the decisions of thrce differcnt clessifiers: template matching using storcd prototypes, a mixed approach thrt uses statistical and slructural analysis of digit boundrry and a rulebased approach to analyze dlgit strokes. The rcsult of ZIP Code rccognltlon is verifled using a postal dircctory. Performlnce of lhe system, stlll under rcfinement, is promising.