A Comparative Analysis And Classification Of Telugu Characters Extracted From Palm Leaves

The challenge of recognizing handwritten characters from palm leaf document image is crucial for digitizing old manuscripts. To accurately identify characters from palm leaf images, this research explores various network designs, including Convolutional Neural Network (CNN), VGG16, MobileNet, and InceptionNet. The cornerstone of this study is the CNN architecture, which captures local characteristics and spatial hierarchies. Additionally, VGG16, with its greater depth, and MobileNet, with reduced complexity, are employed. Experimental assessments reveal that CNN achieves an accuracy of 98.98%, while VGG16 achieves 88.49% accuracy and MobileNet achieves 97.19% accuracy. By utilizing different kernel sizes and pooling techniques, Inception Net achieves an accuracy rate of 97.63%.