RNN-Based Sanskrit Character Recognition Using the NHCD Dataset

Handwritten Sanskrit character recognition is a challenging task due to variations in writing styles, overlapping strokes, and image distortions. This study proposes a Recurrent Neural Network (RNN)-based model optimized for the automatic identification of Sanskrit characters from the National Handwritten Character Dataset (NHCD), which includes vowels, consonants, and numerals. The model leverages Long ShortTerm Memory (LSTM) and Gated Recurrent Units (GRU) to effectively capture spatial dependencies in character structures. To improve recognition performance, the system applies image preprocessing methods like resizing, normalization. The model is trained on a well-structured dataset and assessed using essential metrics such as accuracy, precision, recall, and F1score. Experimental findings indicate the model's strong effectiveness, attaining an accuracy of 94.76% accuracy, showcasing its reliability in Sanskrit character recognition. Furthermore, a relative assessment with CNN, RCNN, and Transformer-based models highlights the strengths of RNN in recognizing sequential patterns in handwritten characters. This research contributes to the development of automated Sanskrit OCR systems, facilitating the digitization and preservation of ancient texts.