Syllabary Pattern Recognition and Enhanced Syllabary Analysis Using Deep Neural Networks from Palaeographic Writings

From the seventh to the tenth century, historical Indian rural inscriptions are challenging but valuable materials for historical study. These texts, inscribed on sturdy materials like stone, offer timeless insights into India's cultural legacy. However, because these inscriptions contain compound letters and variations in the same symbols, contemporary technology like handwriting and optical character recognition encounter substantial challenges when attempting to understand them effectively. Using a Convolutional Bidirectional Long Short-Term Memory (Conv Bi-LSTM) model for character identification and a Syllabary Pattern Recognition (SPR) strategy for syllabary segmentation, this research suggests a unique method to improve the recognition of Tamil paleographic inscriptions. Preprocessing, Convolutional Neural Networks (CNNs) for feature extraction, and LSTM networks for character-level recognition are the steps in the methodology. Experimental results demonstrate an impressive character recognition accuracy of 97.9%, outperforming traditional methods. This research aims to contribute to digitizing and preserving India's cultural heritage by offering a robust solution for automatically recognizing intricate paleographic scripts.