CSM-based feature extraction for degraded machine printed character recognition

Abstract

This paper presents an OCR method for degraded character recognition applied to typewritten document produced by typesetting machine. The complementary similarity measure method (CSM) is a well known classification method and widely applied in the area of character recognition. In this work the CSM method is not only used as a classifier but also introduced as a feature extractor, and applied to degraded character recognition. The resulted CSM feature vector is used to train a multi layered perceptron (MLP). The use of the CSM as a feature extractor tends to boost the MLP and makes it very powerful and very well suited for rejection. Experimental results on n typewritten A4 page documents show the ability of the model to yield relevant and robust recognition on poor quality printed document characters.