Title: Handwritten Odia numeral recognition using combined CNN-RNN

Authors: Abhishek Das; Mihir Narayan Mohanty

Addresses: Department of Electronics and Communication Engineering, ITER, Siksha 'O' Anusandhan (Deemed to be University), Bhubaneswar, Odisha, India; School of Information Technology, SRM University Sikkim, Gangtok, Sikkim, India ' Department of Electronics and Communication Engineering, ITER, Siksha 'O' Anusandhan (Deemed to be University), Bhubaneswar, Odisha, India

Abstract: Detection and recognition of handwritten characters play a vital role in natural language processing. In this work, authors have taken an approach for Odia handwritten numbers recognition. A little work has been done in Odia numeral recognition. IIT-Bhubaneswar Odia handwritten numerals dataset is considered in this work that contains 5164 images. Deep learning models need a large number of data for training. 1000 images generated through Deep Convolutional Generative Adversarial Network (DCGAN) were added to the dataset to increase its size. Since single models could not perform well in classifying the images, a combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) cells are considered for classification. The Adam Optimisation algorithm is used for minimising the error, and to train the network the supervised learning method is used. The proposed deep learning-based hybrid model provided 98.32% recognition accuracy that was found to be competitive with the state-of-art methods.

Keywords: character recognition; Odia numerals; deep learning; CNN; RNN; LSTM; DCGAN; Adam optimisation; hybrid model; numeral recognition; image classification; image generation.

DOI: 10.1504/IJGUC.2023.132619

International Journal of Grid and Utility Computing, 2023 Vol.14 No.4, pp.382 - 388

Received: 06 Feb 2020
Accepted: 24 Jul 2020
Published online: 31 Jul 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article