Handwritten Character Recognition Using CNN

Abstract

In this project I developed a system based on machine learning that helps to recognize the handwritten characters. The idea for this project came from the paper ‘Adith Narayan and Raju Muthalagu, “Image Character Recognition using Convolutional Neural Networks”, 2021 Seventh International conference on Bio Signals, Images, and Instrumentation (ICBSII)’. That paper suggests using Convolutional Neural Networks for developing such a system. This paper aims the study and implementation of Convolutional Neural Network (CNN) for Image character recognition. Handwritten Character Recognition involves recognition of texts present in digital images and documents and processing them for various applications such as machine translation, pattern recognition and so on. This paper studies the use of CNN in detecting and recognizing handwritten text images with a higher accuracy. The CNN model is tested on English handwritten characters and validated on its performance. The model performs feature extraction from images through multiple layers. These are later used for training the model and thereby recognizing characters. I also referred another paper ‘Shah Nawaz, Alessandro Calefati, Nisar Ahmed and Ignazio Gallo, “Hand Written Characters Recognition via Deep Metric Learning”, 2018 13th IAPR International Workshop on Document Analysis Systems’ which uses Deep Metric Learning to deal with the same problem. Deep metric learning plays an important role in measuring similarity through distance metrics among arbitrary group of data. From the two, I chose the method of CNN to develop the project. So here I am using Alexnet architecture of CNN for recognizing handwritten characters. The characters can be uppercase and lowercase English alphabets and digits (0-9). There are 62 classes in all. To train the system, I created my own dataset by merging the dataset from kaggle and those collected from my friends. The dataset includes 6200 images in ‘.png’ format, 100 for each of the 62 classes. The user can upload an image of a single character written through the interface. And by a button click the prediction will be