Clustering based image binarization in palm leaf manuscripts

Abstract

Palm leaf manuscripts are one of the earliest forms of written media that has enlightened humanity with various subjects such as medicine, astronomy, mathematics and astrology. Many palm leaf manuscripts are approaching the end of their natural life time and are undergoing rapid degradation. The primary objective of image processing with such degraded palm leaf manuscripts is to retrieve and preserve the historical knowledge. The main objective of the paper is the accurate extraction of foreground information from palm leaf images. We apply the concept of clustering in palm leaf image binarization with three dimensional features. To demonstrate the usefulness of the proposed method, a set of ground truth corresponding to ten palm leaf images is generated. It allows setting the benchmark for the proposed and existing technique's practical effectiveness. The proposed clustering based method is observed to achieve higher binarization accuracy in palm leaf manuscripts than the thresholding based approaches.