Computer Science Department, ATSS College of Business Studies and Computer Application, Chinchwad, Pune-19

Abstract

Preservation of cultural material, especially historical palm-leaf manuscripts in Kannada, is necessary for understanding the historical, literary and theological transformation of South India. Categorizing these texts and at-tributing their authorship using traditional methods is a time consuming, difficult task. Proposed work is an attempt to apply deep learning technology for author classification of the Kannada Handwritten palm leaf manuscripts. To our knowledge, in this paper we propose for the first time a Convolutional Neural Network ensemble specific to historical Kannada palm leaf manuscript classification, which is a novel approach in heritage document analysis. The study also considers VGG16, DenseNet, AlexNet and multilayer perceptron (MLP) methods. An approach that combines them is considered as the optimal choice of approach because it provides better accuracy and reliability. This proposed methodology of recognizing Kannada palm leaf writings and identifying the author had shown to have a great impact when applied with Deep learning followed by ensemble learning mechanism. We compare these four different methods in terms of accuracy and loss. It is to be noted that AlexNet method achieves an impressive accuracy of 1.00 and a negligible loss of 0.000007, whereas the multilayered method presents approximately lower accuracy of 0.98 with higher value for the loss which are preserved over various epochs in the proposed approach. VGG16 and DenseNet techniques achieve 0.99 accura-cy, but the estimates for their loss are different. We achieve an ensemble accuracy of 1.00 and a minimum loss of 0.000006. The experiment results show that the deep learning models are effective in extracting palm leaf texts. The ensemble methods have reported satisfactory performance. The proposed approach increases automation and categorization of cultural artifacts.