Abstract:
Handwritten character recognition is a field that falls under the domain of image
recognition. It has been under research for years. The main purpose of handwritten character
recognition is to recognize characters written by humans in a paper that is available in digital
form. This research work is focused on recognition English characters including uppercase,
lowercase and the digits using a convolutional neural network.
In this research work, a customized convolutional neural network model is proposed
called E-Character Recognizer after several experiments on different parameter values ofthe
convolutional neural network. The English character dataset, EMNIST is used to test the
performance of E-Character Recognizer which is compared with the different pre-trained
models including VGG-16, VGG-19, DenseNet-121, ResNet50 V2 and Mobile Net V2 on
the same dataset. The problem encountered in the model was confusion due to the similarity
ofthe structures ofsome of the characters like “1” and “I” etc. it has proved to be the main
reason for confusion for the model.
Upon the comparison, the accuracy of the E-Character recognizer is the best as
compared to the pre-trained models. E-Character recognizer has produced better results in
terms of both the accuracy and the training time. The E-Character recognizer has performed
better as compared to the pre-trained model with an accuracy of 87.31 %. The research was
conducted on the Google Colab GPU service.
