Investigation on deep learning for handwritten English character recognition

Handwritten character recognition mostly depends on human intervention and consumes more time. Optical character recognition is predominantly used in many applications and vast amount of research is decisive in this domain for the past one decade. Deep Convolutional Neural Networks (DCNN) is used to solve optical character recognition problem and have been shown to provide significantly better results the conventional methods. In this proposed work hyper parameters like kernel size, padding, stride size, activation function and number of layers will be tuned to improve the CNN-based handwritten English alphabet recognition. A various combination of learning parameters in LeNet-5 architecture is implemented to achieve a better result in classifying MNIST handwritten English characters. LeNet-5 architecture showed better performance in terms of both training and testing accuracy.