An adaptive foreground-background separation method for effective binarization of document images

Abstract

Binarization is a process of classifying the pixels of an image as either foreground or background. Most of the binarization techniques suffer from the noise appearing in the images during acquisition such as uneven illumination. In the present work, a foreground-background separation method is developed to enhance the performance of a document image binarization method. To examine its effectiveness, it is combined with two state-of-the-art binarization methods (i.e. Otsu’s method [1] and Mitianoudis’ method [2]) and the performances of the combined methods are compared with the original methods. For the experiment, two standard databases viz., DIBCO 2012 and 2013 are used. The results confirm that the proposed method performs satisfactorily even if the images are considerably noisy.