Hand-Written Bramhi Script Recognition using Deep Learning

This work involves the classification of handwritten script using deep learning with a special interest in the comparison of different pre-trained models. As one of the oldest writing systems in the world, derived from the Dravidian and southeast Semitic scripts that were developed 4,500 years ago in India and Central America, Brahmi script also has variability due to the hand-written accented forms. To address these obstacles, different pre-trained models viz., Vision Transformer (ViT), MobileNetV2, VGG16, ResNet50, DenseNet, and EfficientNet with the CNN are implemented and compared. The result of the implementation infers that all of these pre-trained models yield a higher accuracy than the custom CNN traditional CNN approach, with a peak improvement in performance shown by MobileNetV2. This lays the ground for future utilization of deep learning architectures which have the capability of preserving and understanding handwritten symbolic scripts.