Handwritten Address Interpretation System Allowing for Non-Use of Postal Codes and Omission of Address Elements

Abstract

This paper reports a new address interpretation system which allows both non-use of postal codes and omission of address elements, such as the omission of county or state when a city name has been given. While lexicon-driven recognizers show good recognition performance when they are input with adequate word images and a lexicon containing correct word strings, it is difficult to design one which would be of practical use when postal codes are not in use and elements of addresses may have been omitted. That is to say, an inadequate design in this area is likely to result in an impractically high erroneous recognition rate. In response to this problem, we propose here an advanced address interpretation system that utilizes both an improved address interpretation method and improved word recognition methods. The improved address interpretation method has been designed to satisfy as completely as possible the need to accommodate non-use of postal codes and the omission of address elements, while the improved word recognition methods have been designed to achieve low erroneous recognition rates in cases in which that need has not been fully satisfied. When we applied our new system to approximately 2000 actual address images for which that need would be relevant, we achieved a ¾± rate of correct outward sorting with only a ¼ ± rate of erroneous outward sorting. These rates are good enough for practical applications.

Handwritten Character Recognition on Focused on the Segmentation of Character Prototypes in Small Strips

“…scanned document). The importance and need of handwriting recognition has been arising in many real world applications such as postal addresses [1], bank checks processing [38], handfilled forms processing [29], conversion of field notes and historical manuscripts where a vast amount of handwriting historical manuscripts can be easily converted into an electronic representation for sharing and searching [24].…”

SwiftPost — A vision-based fast postal envelope identification system

“…A typical automatic postal address recognition system mainly consists of five modules: (1) envelope image preprocessing, (2) address block location, (3) line/character segmentation, (4) character recognition, (5) address post-processing. Throughout the intensive research of last three decades, a great deal of improvement has been achieved in each sub-topic and several systems for postal addresses recognition are also available [2][3][4][5][6][7][8][9][10][11][12]. However, most of these works were mainly focused on handwritten address recognition.…”