Article’s

This paper presents a real-time handwritten character recognition system capable of identifying handwritten English alphabets (A-Z) and numeric digits (0-9) using deep learning. The system employs two separate Convolutional Neural Network (CNN) models — one trained on the A-Z Handwritten Character Dataset comprising 372,450 samples and another trained on the MNIST dataset comprising 60,000 digit samples. A confidence-based router mechanism determines whether a drawn character is a letter or digit by comparing output probabilities of both models and selecting the prediction with higher confidence. Both models achieve 99% validation accuracy after 5 training epochs. A real-time graphical user interface built using Tkinter allows users to draw characters on a digital canvas and receive instant predictions. Key contributions include an auto-crop preprocessing technique that eliminates scale sensitivity, a dual-model confidence router for letter-digit disambiguation, and contour-based character segmentation for word and sentence level recognition.