Manuscripts Character Recognition Using Machine Learning and Deep Learning

The automatic character recognition of historic documents gained more attention from scholars recently, due to the big improvements in computer vision, image processing, and digitization. While Neural Networks, current state-of-the-art models used for recognition, are very performant, they typically suffer using large amounts training data. In our study we manually built own relatively small dataset 404 characters by cropping letter images a popular manuscript, Electronic Beowulf. To compensate use ImageDataGenerator, Python library was augment Beowulf manuscript’s dataset. augmented once, twice, thrice, which call resampling 1, 2, 3, respectively. classify efficiently, developed customized Convolutional Network (CNN) model. We conducted comparative analysis results achieved proposed model with other machine learning (ML) such as support vector (SVM), K-nearest neighbor (KNN), decision tree (DT), random forest (RF), XGBoost. pretrained VGG16, MobileNet, ResNet50 extract features images. then trained tested above ML recorded results. Moreover, validated CNN against well-established MNIST Our achieves good accuracies 88.67%, 90.91%, 98.86% cases respectively, Additionally, benchmark accuracy 99.03%