Abstract
Deep learning based algorithms are used in various pattern recognition tasks, including character recognition. Convolutional Neural Network (CNN) is effectively implemented for character recognition and is one of the best performing deep learning models. CNN can be used for character recognition directly or it can be used for extracting features in the character recognition process. Implementation of a feature extraction method using CNN autoencoder for MODI script character recognition is discussed in the paper. The extracted features are then subjected to Support Vector Machine (SVM) for the purpose of classification. The On-the-fly data augmentation method is used to add variability and generalization of the data set. MODI Script is an ancient Indian script and was used for writing Marathi until 1950. Various libraries and temples in India and abroad have a large collection of MODI documents. Character recognition related research of MODI script is still in infancy and research and development is necessary to extract the information from MODI manuscripts stored in various libraries. The performance of the proposed method, which uses CNN autoencoder as a feature extractor and an SVM based classifier gives very high accuracy and is better compared to the most accurate MODI character recognition method reported so far.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have