Scene Text Recognition of Indian Languages in Natural Scene Images

Scene Text Recognition (STR) for Telugu and Sanskrit is challenging due to complex character shapes, modifiers, and noisy backgrounds in natural images. Traditional OCR systems struggle to handle such variations effectively. This paper proposes a hybrid deep learning model that combines Convolutional Neural Networks (CNN) for spatial feature extraction with Bidirectional Long Short-Term Memory (Bi-LSTM) networks for sequence learning. The system is trained on a large labeled dataset with preprocessing techniques such as resizing and normalization. Experimental results show improved accuracy and robustness, making the approach suitable for digitization and assistive applications.