Knowledge-based contextual processor for text recognition

Abstract

This paper describes a knowledge-based system for the recognition of text found in medical books. It serves as a smart front-end for a bigger project on automatic knowledge acquisition. The focal point of the system is a knowledge-based contextual processor that uses lexical, syntactic and semantic information to improve the performance of the overall system. The natural language processing rules are used immediately after a word is recognized to provide feedback to the classifier during a run. The idea is to generate possible candidates from the input word and then filter out unlikely ones so that the most likely word is selected at the end. Currently, we are able to achieve a recognition accuracy of above 995%. We expect the recognition accuracy to be better when the entire system is completed.