CNN based Recognition of Handwritten Mathematical Expression

Abstract

Handwritten mathematical expression recognition remains a critical challenge in digitized computation due to variations in writing styles and noise in input data. This research addresses the gap in accurate recognition and evaluation of multi-operator handwritten expressions by integrating a convolutional neural network (CNN) with an optimized preprocessing pipeline. The proposed system processes both user-drawn and uploaded handwritten expressions via a web-based interface. The backend, developed using Python and Flask, employs grayscale conversion, bit inversion, binary thresholding, and contourbased cropping to enhance feature extraction. The CNN, trained on preprocessed grayscale images of digits and operators, predicts mathematical expressions, which are subsequently evaluated for accuracy. Experimental results demonstrate high recognition precision and efficient computation of complex expressions. The findings contribute to advancing automated mathematical interpretation, facilitating applications in education and professional domains by improving accessibility, usability, and computational efficiency.