HMER-Image To LaTeX : A Variational Dropout Approach

Abstract

A lot of research has been done to convert textual information into digital format but cannot obtain the desired results when applied to handwritten mathematical expressions. It is mainly due to complex 2D structure and spatial relations of the expressions. Meanwhile, many models were developed using encoder-decoder-based approaches to solve the problem. In this paper, we present a novel approach based on an encoder-decoder model with variational dropout. The model takes an image of a handwritten mathematical expression as input and provides the one-dimensional LaTeX code as output. It achieved 35.8% and 34.6% accuracy on CROHME 2014 and 2016 datasets, respectively.