Deep Learning Based Enhanced Handwritten Devanagari Character Recognition using Image Augmentation
Khushi Sinha, Eshan Marwah, Richa Gupta · 2023
This research work aims to improve the accuracy in identifying characters written in the Devanagari script, which is extensively used in a number of South Asian languages, including Hindi, Sanskrit, and Nepali, etc. The proposed work includes the development of a Convolutional Neural Network (CNN) model for enhanced handwritten Devanagari character recognition, with an emphasis on enhancing model performance through data augmentation. Data augmentation involves a set of techniques that increase the size of the dataset and balance classes by applying various transformations, aiding model prediction, and overfitting. Results are also analyzed on the basis of different performance measures like accuracy, precision, recall and f1-score. The proposed CNN model provides high accuracy with batch normalization and dropout layers, improving model training and generalization.