부산대학교 도서관

Handwritten Character Recognition (HCR) has gained significant popularity as an Artificial Intelligence tool in today's world. Recognition of a particular character is one of the challenging tasks when every work is done digitally. This technology converts handwritten characters to machine readable or digital form by applying various machine algorithms. Since handwriting and the style of writing a particular character always varies from one person to another person. A person may write some character whose style may differ in size, shape, font and position. An implementation a Handwritten Character recognition technology that uses deep learning algorithm is used. The dataset used in this technology is EMNIST (Extended Modified National Institute of Standards and Technology) dataset. The format of this dataset is of CSV file and it contains handwritten characters, which includes both uppercase and lowercase letters along with digits and various symbols. The dataset used in this model is divided into two parts one is training and other is testing. We have performed an operation for reshaping the image into 28 x 28 pixels so that it can be fitted with Convolutional Neural Network (CNN). During the training process training data is iterated through multiple epochs. The accuracy of the model is measured simultaneously. During the testing process several grayscale images are used to test the trained model and the prediction is made accordingly This model presents the accuracy, potential and high performance of CNN model in prediction of a handwritten character. This system opens up possibilities for its usage in various other applications, as it speeds up our task of analyzing documents and digitization.