Handwritten Character Recognition on Android for Basic Education Using Convolutional Neural Network

Abstract

An international initiative called Education for All (EFA) aims to create an environment in which everyone in the world can get an education. Especially in developing countries, many children lack access to a quality education. Therefore, we propose an offline self-learning application to learn written English and basic calculation for primary level students. It can also be used as a supplement for teachers to make the learning environment more interactive and interesting. In our proposed system, handwritten characters or words written on tablets were saved as input images. Then, we performed character segmentation by using our proposed character segmentation methods. For the character recognition, the Convolutional Neural Network (CNN) was used for recognizing segmented characters. For building our own dataset, handwritten data were collected from primary level students in developing countries. The network model was trained on a high-end machine to reduce the workload on the Android tablet. Various types of classifiers (digit and special characters, uppercase letters, lowercase letters, etc.) were created in order to reduce the incorrect classification. According to our experimental results, the proposed system achieved 95.6% on the 1000 randomly selected words and 98.7% for each character.