Evaluation of Binarization Methods for Aged Printed Myanmar Documents

Abstract

Binarization is one of sub phases of preprocessing step of optical character recognition (OCR). Binarization is separation of foreground text from background of document image. The accuracy of OCR mainly relies on binarization’s result. This paper compares several alternative binarization algorithms for aged printed Myanmar documents. The algorithms evaluated are global thresholding (Otsu), Local thresholding (Niblack, Sauvola, Wolf, Feng and Nick). It is found that the binarized images more stable if filters (Wiener and Gaussian) are prior used before applying binarization algorithms. Another one is that local thresholding is suit for aged Myanmar documents. Among local thresholding, Niblack, Sauvola and Wolf are the more suitable algorithms based on the experimental results. The quality of binarized images is verified by using different assessment parameters like mean square error (MSE), signal to noise ratio (SNR) and peak signal to noise ratio (PSNR). This work aims to get the high accuracy of recognition steps with the main objective of developing OCR of aged printed Myanmar documents.