Recognition of Handwritten Mathematical Expressions Using Systems of Convolutional Neural Networks

Abstract

Accurate recognition of handwritten mathematical expressions has proven difficult due to their two-dimensional structure. Various machine-learning techniques have previously been employed to transcribe handwritten math, including approaches based on convolutional neural networks (CNNs) and larger encoder/decoder-based models. In this work, we explore a CNN-based method for transcribing handwritten math expressions into the typesetting language known as LaTeX. This approach utilizes machine learning not only for classifying individual characters but also for extracting individual characters from handwritten inputs and determining what forms of two-dimensionality exist within the expression. This approach achieves significant reliability when recognizing common mathematical expressions.