Region Based Local Binarization Approach for Handwritten Ancient Documents

Due to the fact that historical handwritten documents present many degradations, pre-processing of such documents is considered as a big challenge. Most pre-processing methods and specifically binarization return better results when they are applied on printed documents. We present in this paper a binarization approach adaptive for handwritten historical documents based on extraction of regions-of-interest. During our tests several images datasets are used, the benchmarking datasets for binarization DIBCO 2009 and H-DIBCO 2010 (15 images) as well as complete handwritten documents from the IAM historical database (about 60 images). The evaluation of the proposed binarization method is based on several evaluation metrics for binarization. The results show that the proposed method fit with handwritten historical documents (FM about 88%) for images of the binarization competitions.