A CNN-Based Approach for Handwritten Mathematical Formula Recognition and LaTeX Generation

The recognition of handwritten mathematical formulas is a challenging task due to the complexity of mathematical symbols, varied notations, and diverse handwriting styles. This paper introduces an innovative Handwritten Mathematical Formula Recognition System (HMFRS) that leverages deep learning techniques to achieve high accuracy in recognizing handwritten formulas. The system comprises three key modules: pre-processing, feature extraction, and formula recognition. In the pre-processing stage, the handwritten formula image is enhanced to improve clarity and eliminate noise. The feature extraction module utilizes Convolutional Neural Networks (CNNs) to transform the pre-processed image into a feature representation, capturing both the local and global structure of the formula. We employ a comprehensive dataset consisting of handwritten mathematical formulas across different writing styles and symbol sets to train and evaluate the HMFRS. Experimental results demonstrate that our system outperforms existing methods in terms of accuracy and robustness. The proposed system holds promise for applications in areas such as digital document processing, e-learning platforms, and scientific research, where accurate recognition of handwritten mathematical formulas is essential. Additionally, we focus on recognizing and solving handwritten quadratic equations, employing horizontal compact projection analysis and combined connected component analysis for segmentation, followed by CNN-based character classification. The results validate the effectiveness of our approach in providing accurate solutions for mathematical equations.