Evaluation of Handwritten Arithmetic Equations using Convolution Neural Networks


  • Chandra Sekhar Sanaboina, Rahul Sidda, Ahmed Akram, Sruthi Samuel Sarella, Jennifer Rogers


This paper aims to develop a user interactive assistance system that evaluates the handwritten arithmetic equations based on handwriting recognition algorithms. Although recognizing handwritten characters and symbols is generally easy for anyone but recognizing them is difficult for a machine. By following a deep learning approach, this challenge can be solved by designing a system that recognizes the operands and operators. Being able to solve handwritten arithmetic equations through the model will bring faster and accurate results. The model will identify pictures of handwritten arithmetic equations and will be able to emit the corresponding characters into a list and evaluates the results.This includes digit classification which involves feature extraction and classification. For this purpose, computer vision is used to input the image and obtain contours, Convolutional NeuralNetwork(CNN) is the algorithm used to build the model, which does feature extraction and classify the operators and operands and develop a model with 92.6% accuracy.
Keywords - Computer Vision (CV), Feature extraction, Contour extraction, Convolutional Neural Network (CNN), TensorFlow, Training model, Keras, ReLu, Softmax.