An Improved Sauvola Approach on Document Images Binarization

Abstract

Document image binarization is one important processing step, especially for data analysis. A variable background, non-uniform illumination, and blur give a big challenging task in order to detect the text. In this paper, a new binarization based on local thresholding technique ‘WAN’ was presented. The proposed algorithm is known as ‘WAN’ after the first name of the author of this paper. WAN has been inspired by the Sauvola’s binarization method and exhibits its robustness and effectiveness when evaluated on low quality document images. Sauvola method failed to segment if the contrast between the foreground and background is small or if the text is in thin pen stroke text. The objective of the WAN method is to improve the Sauvola method and achieved a better binarization result. The results of the numerical simulation indicate that the WAN method is the most effective and efficient (f-measure 72.274 and NRM = 0.093) compared to the Sauvola method, Local Adaptive method, Niblack method, Feng Method, and Bernsen method.