Abstract

Vatteluttu (Vattezhuthu) script, prevalent between the 3rd and 8th century CE, played a crucial role in documenting early Chola and Pandya history. However, despite its historical significance, thousands of inscriptions remain untranslated, hindering a comprehensive understanding of early South Indian heritage. This research addresses these challenges by developing a deep learning-based approach for digitizing and recognizing Vatteluttu characters. Using a dataset of 1,800 segmented images representing 28 characters, the study integrates advanced preprocessing techniques and a Siamese CNN-RNN architecture to classify ancient scripts with an overall accuracy of 98%. The findings demonstrate the feasibility of automated transcription for ancient scripts, offering a robust framework for preserving and enhancing research on South Indian heritage.