Computer Recognition of Hand-printed Text

Abstract

This paper describes the procedures and results of a project aimed at the computer recognition of relatively unconstrained hand-printed texts. Scanning, feature-extraction, and character classification techniques are described. Novel approaches investigated include a feature-extracting preprocessor consisting entirely of local edge detectors, the use of multiple-valued confidence indicators both before and after classification, the combining of independent preprocessor-classifier systems in parallel to achieve improved character-recognition accuracy, and the use of application-oriented context analysis. Two large files of hand-printed data are described, and results concerning their legibility are given. An extensive bibliography in hand-printed character recognition is included.