Offline Handwritten Mathematical Expression Recognition using CNN and Xception

Abstract

Mathematical expressions generally play a requisite role in scientific communications. They are not just used for numerical calculations, on the other hand, are also employed for fetching scientific information with less ambiguity, and facilitate researchers to exactly outline and formalize target problems. It takes far longer to manually enter mathematical formulas into a computer than it does to write them down on paper using a pen. Recently, we proposed deep learning methods that can identify images of trigonometric expressions from 2dimensional layouts to 1dimensional strings in order to solve this issue. As densely connected convolutional neural networks (CNN) can boost accuracy, we utilize CNN to improve the results in this study. In order to compare performance, the Transfer Learning framework Exception is employed, which obtains 90% accuracy when recognizing handwritten mathematical expressions. CNN provides 98% accuracy in this regard. Therefore, the CNN model that we created has a higher accuracy rating than the Transfer Learning model Xception