A Hybrid SVC-CNN based Classification Model for Handwritten Mathematical Expressions(Numbers and Operators)

Abstract

Machine learning and Computer Vision are computer science domains that have been working closely for a long time. Given the ubiquity of handwritten text in human transactions, we are endeavoring to acquire the answer to the quest "Can Computer Vision and Machine learning together be deployed effectively for decisive classification of handwritten mathematical numbers and operators?". The easier it is to communicate via handwritten texts and documents, the more challenging the task of digitizing and prediction, especially for the two-dimensional complex math statements and operators. This paper presents a hybrid model that involves machine learning and deep learning-based decision algorithms for classifying and predicting mathematical numbers and operators. The dataset considered for the experimentation has been downloaded from the Kaggle dataset store consisting of more than 12K images. The primary tasks involved include data collection, data preprocessing, and building and deploying the model. Mainly our model focuses on the extraction of contour features and performing classification using the LinearSVC model, and the prediction of numbers has been accomplished using CNN. The proposed classification and prediction model achieves an accuracy of 89.76% for predicting the math operators and 91.48% for predicting the numbers.