Optical Character Recognition of Sanskrit Manuscripts using Convolution Neural Networks

Optical Character Recognition of Sanskrit Manuscripts Using Convolution Neural Networks delves into the cutting-edge application of deep learning for deciphering Sanskrit manuscripts written in Devanagari script. Tackling one of the most challenging tasks in OCR—recognizing Sanskrit’s intricate characters and symbols—this work presents a robust system designed to enhance recognition accuracy for scanned text images. By employing advanced architectures such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM, alongside traditional classifiers like k-Nearest Neighbors (KNN) and Support Vector Machines (SVM), the research achieves remarkable accuracy rates. Beyond single-touching characters, it innovatively addresses overlapping lines, connected letters, and half-characters, providing solutions to limitations in existing systems. With a peak recognition accuracy of 98.64% for mixed Sanskrit text, this study is a vital contribution to the preservation and digitization of ancient literature. It opens new doors to computational linguistics, ensuring Sanskrit's cultural heritage thrives in the digital age.