Offline Recognition of Handwritten Mathematical Expression using Neural Networks

The identification of handwritten mathematical expressions is critical in many fields, including education, engineering, and scientific research. This study provides a complete overview and analysis of cutting-edge approaches and procedures used in the development of handwritten mathematical expression recognizers.This work introduces a high-performance HMER model with scale augmentation to improve the recognizer’s robustness to writing styles. The attention-based encoder-decoder network in our proposed model is specially designed to extract features and provide predictions using Scale Augmentation. Scale augmentation notably improves the model’s performance on MEs with variable scale in both the horizontal and vertical axes. When the decoder’s attention allocation is poor, dropping attention is advised to improve performance even further. To demonstrate the usefulness of each component of the method, we conducted extensive testing on the CROHME dataset and obtained high accuracy on the 2014, 2016, and 2019 datasets.