Recognition of Simple Handwritten Polynomials Using Segmentation with Fractional Calculus and Convolutional Neural Networks

Abstract

This work introduces a method for recognizing handwritten polynomials using Convolutional Neural Networks (CNN) and Fractional Order Darwinian Particle Swarm Optimization (FODPSO). Segmentation of the input image is done with the FODPSO technique, which uses fractional derivative to control the rate of particle convergence. After segmentation, three CNN are used in the character recognition step: the first one classifies the individual symbols as numeric or non-numeric. The second network recognizes the numbers, while the third CNN recognize the non-numeric symbols. A heuristic procedure is used to build the polynomial, whose graph is finally plotted. A total of 264780 images containing symbols and numbers were used for training, validating, and testing the CNN, with an accuracy of approximately 99%.