Details

This project develops a handwritten character recognition system using a Convolutional Neural Network (CNN) with semi-supervised learning to improve accuracy through pseudo-labeling. A confidence-based filtering mechanism ensures only high-confidence predictions contribute to training, reducing noise and enhancing generalization. The modular design allows experimentation with Vision Transformers, RNNs, and LSTMs, while an interactive GUI enables real-time character input with instant feedback. Trained on the MNIST dataset, the model is adaptable for other datasets and optimized for potential edge deployment on mobile and low-power devices, making it a versatile and practical AI solution.