Binarization of historical documents using self-learning classifier based on K-Means and SVM

This article aims to present a new binarization method of degraded historical document images. The new algorithm combines K-Means classification with a classical binarization method to generate a pure learning set and a conflict class. We use SVM classifier to manage the conflict class in order to make the final binarization that classifies each pixel of image document as foreground or background. Experiments are conducted on the standard datasets Dibco 2009 and Dibco 2011. The obtained results are very promising that allows opening a large margin of investigation.