Hand Written Characters Recognition Using Deep Learning

Abstract

This paper provides a new solution to conventional techniques for handwriting recognition using Deep learning ideas and computer vision. An expansion of MNIST digits dataset was used, called the Emnist dataset. This contains 62 classes in both upper and lower case with 0-9 digits and A-Z characters. An application has been developed for Android to identify handwritten text and transform it to digital form using Convolutional Neural Networks, abbreviated as CNN, to identify and detect text. We pre-processed the dataset before that, and used different filters to it. Using Android Studio we created an android application and connected our text recognition program to handwriting utilizing tensorflow libraries. The application's layout was kept simple for display purposes. The experienced keras graph is used through a protobuf file and tensorflow interface to anticipate alphanumeric characters drawn with a finger.