IMPROVING OPTICAL CHARACTER-RECOGNITION ACCURACY USING ADAPTIVE IMAGE-RESTORATION

A technique that can improve the optical character recognition (OCR) a ccuracy of text images is presented. By using the output from an OCR s ystem and a distorted text image, this technique trains an adaptive re storation filter and then applies the filter to the distorted text tha t the OCR system could not recognize. The restored text image is then reprocessed by the OCR system, and the restored characters are recogni zed with a higher accuracy than the distorted text A series of experim ents were performed to determine a feasible adaptive restoration filte r architecture, to establish this technique's ability to restore unifo rmly distorted text, and to demonstrate this technique's ability to im prove the OCR accuracy of real world text documents. The results of th ese experiments show that this technique can improve both pixel and OC R accuracy of distorted text images. (C) 1996 SPIE and IS&T.