Deep Learning Algorithms for Recognizing and Digitizing Scanned Kannada Printed and Handwritten Documents

In this research, we explore the recognition of scanned and handwritten Kannada characters using machine learning, specifically employing the Inception-v3 model. Unlike static printed documents, digitized handwritten texts can be edited, This is essential in several applications like speech conversion, language translation, and the conservation of historical documents. Kannada, a language with a complex script, presents a challenge for text recognition, particularly when dealing with handwritten characters. The Inception-v3 The model was selected since it can improve feature propagation, enhance feature reuse, as well as lower the vanishing gradient problem. We tested the model on two datasets-kannada-ocr-test-images-with-ground-truth and Scribd - Labour Law 1 - Hand Written Notes Kannada in A4 Sheets-to recognize scanned and handwritten Kannada characters. The results showed a testing accuracy of 83.1% for scanned printed images and 86.7% for scanned handwritten images. For comparison, the Tesseract OCR tool, a widely-used standard OCR system, was also tested, and while it performed well on both types of images, Inception-v3 outperformed it in precision, recall, and F1-score. This demonstrates the superior performance of deep learning models like Inception-v3 over traditional OCR tools in recognizing complex handwritten scripts. The suggested system offers an extensive amount of opportunity for real-world applications, especially in automating the digitization of Kannada administrative documents and preserving historical texts. Additionally, it could be extended to other Indian languages, further enhancing its applicability in multilingual regions of India.