Author ORCID Identifier
Orcid ID: https://orcid.org/0009-0003-1110-6142
Biosketch
Dr. P. Hariharan received the Bachelor of Engineering (B.E.) degree in Computer Science and Engineering in 2010 and the Master of Science (M.S.) degree in Computer Science and Engineering from the Northumbria University-United Kingdom in 2013. He completed his Doctorate (PhD) in 2026 at SASTRA Deemed University, Thanjavur, where his research focused on advanced artificial intelligence methodologies for the recognition and preservation of ancient Tamil palm leaf manuscripts. He is currently serving as an Assistant Professor in Department of CSE from School of Computing (SoC) at SRM Institute of Science and Technology, Trichy Campus (621105). His academic and research contributions are centred on the development of multimodal deep fused neural architectures for Tamil palm leaf manuscript recognition and restoration. He has made significant contributions in applying advanced deep learning and image segmentation techniques to the digitization, analysis, and preservation of historical manuscripts, thereby supporting the conservation of Tamil cultural heritage through modern computational intelligence approaches. His broader research interests include deep learning, machine learning, natural language processing, computer vision, image processing, and intelligent document analysis. Through his interdisciplinary research, he continues to contribute toward innovative AI-driven solutions for complex pattern recognition and heritage informatics applications.
Date of Award
30-3-2026
Document Type
Thesis
School
School of Computing
Programme
Ph.D.-Doctoral of Philosophy
First Advisor
Dr.N.Sasikaladevi
Keywords
Deep Learning, Machine Learning, Image Processing
Abstract
Digitizing Tamil palm-leaf manuscripts is important for education, communication, and the preservation of cultural heritage. The complex structure of the Tamil script, the wide range of handwriting styles, and the degradation seen in ancient Tamil palm-leaf manuscripts make these texts very difficult to read and understand. Digital Image Processing (DIP), document analysis techniques, and traditional Optical Character Recognition (OCR) are unable to handle noise, background interference, faded ink, and limited labelled data, motivating the need for robust, effective Deep Learning (DL)- based solutions.
As a prerequisite to understanding and designing effective recognition systems for ancient manuscripts, this thesis first examines modern Tamil Handwritten Character Recognition (THCR). A DL framework, ResTANet, based on a modified ResNet-101 architecture, is proposed to learn discriminative features from modern handwritten Tamil characters. Effective preprocessing and feature optimization methods are employed. Experiments conducted on our own dataset, Tamil Isolated Character Dataset (TICD), which is publicly available, demonstrate strong recognition capability, achieving 95.36% accuracy, with precision, sensitivity, and F1-score values of 92.71%, 92.61%, and 92.71%, respectively, establishing a reliable baseline for ancient Tamil OCR.
This thesis further proposes a DL-based restoration model, Res-UNetB, that combines ResNet101 as an encoder with a U-Net as a decoder to address the restoration issues of degraded palm-leaf manuscripts. The model maintains fine structural details and effectively separates foreground text from complex backgrounds. This study uses the Large-scale handwritten document for Image Binarization (LS-HDIB) dataset. The model is also compared with other benchmark datasets, including LS-HDIB, PHIBD, AMADI LONTARSET, and H-DIBCO. The robustness and effectiveness of the Res-UNetB method over conventional methods are shown by the highest F1-scores of 28.5%, 22.2%, 29.2%, and 40.7%, respectively.
Based on the results obtained from ResTANet and Res-UNetB, a Semi-Supervised Deep Neural Architecture (SSDNA) framework is developed for the recognition and digitization of ancient Tamil palm-leaf manuscripts. SSDNA includes image enhancement, contour-based character segmentation, and a deep neural network classifier for recognizing individual characters from degraded ancient Tamil palm-leaf manuscripts. Experiments conducted on original Tamil palm-leaf manuscripts demonstrate consistent, reliable recognition performance. The SSDNA model attains final accuracies of 95.65% during training and 95.20% during validation.
Precision, sensitivity, and F1-score values of 88.37%, 90.58%, and 87.97%, respectively, are obtained, confirming its effectiveness in learning from limited labelled data. In addition, a user-friendly web-based application (Olai Suvadi Reader) has been developed to facilitate practical deployment, automatic restoration, and character recognition of palm leaf manuscripts. Collectively, this thesis contributes significantly to the digital preservation and accessibility of Tamil cultural heritage by developing an efficient and scalable DL framework that combines modern handwritten character recognition, manuscript restoration, and historical Tamil palm-leaf character recognition.
Recommended Citation
P, Hariharan Mr, "Development of Deep Fused Neural Architecture for Ancient Tamil Palm-Leaf Manuscript Recognition" (2026). Theses and Dissertations. 206.
https://knowledgeconnect.sastra.edu/theses/206