Classification of Ancient Epigraphs into Different Periods Using Random Forests

Epigraphists, who identify the ancient inscriptions, reconstruct, translate, draw conclusions about the writings, and classify their uses according to dates, are decreasing in number and also because of the fact that repetitive tasks can be exhausting for humans and prone to errors there is a need arising for the automation of these kinds of tasks. It is observed that the characters of a script have evolved over years and transformed to the current form. The purpose of this work is to estimate the period of an epigraph which is the initial step towards automating the task of reading and deciphering inscriptions. The proposed system considers a reconstructed grayscale image of an epigraph pertaining to ancient Kannada script as its input, which is binarized using Otsu's method and then segmented to characters using Connected Component analysis. Normalized Central Moments and Zernike Moments are extracted from the segmented characters and used as the feature vectors for classification. Random Forest (RF) is used as the classifier, which is an ensemble of classification trees, and each tree votes for a class and the output class is the majority of the votes which determines the era of the input epigraph. The system developed is used to classify ancient Kannada epigraphs belonging to the period of any of these dynasties: Ashoka, Satavahana, Kadamba, Chalukya, Rastrakuta and Hoysala. The system showed good results when tested on 110 Kannada epigraph images from different eras. An analysis of the prediction rate of the epigraphs was carried out and obtained a rate of 85% using RF classifier.