CNN-Based Handwritten Mathematical Symbol Recognition Model

Mathematical expression or formula recognition system is evolving field in pattern recognition. Various methods are implemented by different researchers on wide areas. Handwritten mathematical expression field evolved in last decades rapidly. This paper aims to target and represent CNN as an application of the deep learning method for recognizing the handwritten mathematical symbols. The current trend of the deep learning architectures witnesses CNN is one of the most prominent and widely used techniques that have been successfully implemented in NLP, computer vision, and pattern recognition. The kinds of deep learning models have proven to produce a recognizable state of the art. In this paper, we recognize the handwritten mathematical symbols using CNN. This paper comprises stages involved in recognition process, challenges and proposed methodology used by the author to conduct their experiment and further result and future scope is discussed. Dataset used for this experiment is downloaded from the public available platform (HasyV2 Dataset 2018. A dataset of size 369 classes comprising 168,223 images has been used for experimentation, and the results relieve the accuracy of 76.17%. For further improvement, an extra dense net (fully connected layer) has been used. The DuosdenseNet model has been designed; the accuracy of recognizable results shows a decent work for consideration.