Scalable handwritten digit recognition application using neural network and convolutional neural network on heterogeneous architecture

Abstract

Recognition of handwritten digit is one of the popular problem associated with computer vision applications. The goal of our research work is to develop scalable Neural Network(NN) and Convolutional Neural Network (CNN) model that would be able to recognize and determine the handwritten digits from its image. Capability of developing the new algorithms and improve the existing algorithms is determined by the accuracy and speed factor for training and testing the models. In this context, performance of the GPUs and CPUs for handwritten digit system and effects of accelerating the training models have been analyzed. The training and testing has been conducted from publicly available MNIST handwritten database. Web based, offline and online handwritten digit recognition system is developed by using Convolutional Neural Network.