Multiple classifier for degraded machine printed character recognition Multiple classifier for degraded machine printed character recognition

The general problem of optical character recognition (OCR) remains a fundamental but not entirely solved issue in document analysis. In spite of significant improvements in the area of optical character recognition, the recognition of degraded printed characters, in particular, is still lacking satisfactory solutions. This paper presents an OCR method that combines the Hopfield network with a set of autoassociators for degraded character recognition. In the serial combination, the first classifier must achieve lower errors and be very well suited for rejection, whereas the second classifier must allow only low errors and rejects. A relative distance is used as a quality measurement parameter which makes the Hopfieldbased classifier very powerful and very well suited for rejection. We report experimental results for a comparison of three methods: the Hopfield model, the autoassociator-based classifier and the proposed combined architecture. Experimental results show the ability of the model to yield relevant and robust recognition with no errors on poor quality bank check characters even when the patterns are highly degraded. Mots-clés : Hopfield model, associative memory, degraded printed characters, autoassociator, OCR, serial combination, character recognition..