A hybrid approach for document image binarization

In this digital era, people can handle large amount of digital document images and with this document we need to perform several tasks. But there are many challenges addressed in document images, some of them are faded characters, blurred handwritten manuscripts, ink stains and scratched documents. These leads to create problems to identify the proper information which are available in the document images. This situation raised the concern about the development of new techniques and algorithms which can handle these challenges. Image binarization technique is used to improve the quality of the document image and it converts the document image text information (foreground) as black color and the background as white color. The algorithms used to perform image binarization are Otsu, Niblack, Savola and Nick. This paper proposed a hybrid algorithm which is a combination of Otsu and Niblack for documents image binarization. The proposed and the existing algorithms are tested on a synthetic and tobacco dataset. The proposed hybrid method is compared with global and local thresholding methods using standard measures and it is found that the proposed hybrid method has produced good results.