A novel scheme for handwritten binarization method on sundanese palm leaf document images

Palm leaf, one type of ancient Sundanese writing medium, is presenting novel challenges in document image analysis. There are certain characteristics of Sundanese palm leaf document images offering some suitable challenges for handwritten binarization process such as non-uniform illumination, shadow, smear and random noise. We evaluated the performance of five binarization methods, in both common methods (Otsu, Niblack, Sauvola, Bernsen) and current methods (Howe) using the comparative experimental studies. After investigation, we proposed a new scheme for handwritten binarization process that adds special filtering to remove non-text regions. The results show that our proposed filter can improve the performance of all common binarization processes up to 85%. Niblack method which implements our proposed filter gave the highest performance on binarization process. In this research, the images dataset was obtained from sample images of Sundanese palm leaf manuscripts written in XV-XVII century. The evaluation criteria provided by ICDAR 2013 Document Image Binarization Contest (DIBCO) were used in these experiments.