ORIGINAL RESEARCH article
Front. Comput. Sci.
Sec. Computer Vision
RDC-GAN: Generating Realistic Bengali Handwritten Script using Regularized Deep Convolutional Generative Adversarial Network Approach
Provisionally accepted
- 1Devang Patel Institute of Advanced Technology and Research, Charotar University of Science and Technology, CHARUSAT Campus, Anand 388421, Gujarat, India, Anand, India
- 2Dr. K. C. Patel R & D Centre, Charotar University of Science and Technology, CHARUSAT Campus, Anand 388421, Gujarat, India, Anand, India
- 3Indukaka Ipcowala Institute of Management, Charotar University of Science and Technology, CHARUSAT Campus, Anand 388421, Gujarat, India, Anand, India
- 4B. D. Patel Institute of Paramedical Sciences, Charotar University of Science and Technology, CHARUSAT Campus, Anand 388421, Gujarat, India, Anand, India
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
The generation of good quality handwritten Bengali script faces substantial difficulties because of its intricate nature and the absence of multi-faceted data sources. This study develops RDC-GAN as a custom text script generation platform using a CNN architecture which provides improved textscript quality, enhanced generation stability while dealing with Bengali script complexity limitations. The generator incorporates Conv2DTranspose layers along with LeakyReLU activation for its implementation while using Conv2D layers with feature extraction capabilities within the discriminator component. The Adam optimizer maintains stable training when used with a learning rate of 0.00015 and beta1 value of 0.6. Minibatch Discrimination prevention mode collapse through integration with Spectral Normalization that stabilized gradients while the Two-Time Scale Update Rule (TTUR) adjusted generator-discriminator learning balance. The developed optimizations led to better efficiency and more effective results during training. Three evaluation metrics measured the quality of the generated Bengali handwriting: Fréchet Inception Distance (FID), Kullback-Leibler (KL) divergence, and Wasserstein distance. The model designed realistic Bengali textscript within 10 epochs by reaching 0.23 KL divergence and 0.0044 Wasserstein distance, and a 5.133 FID score using generator (6.7048) and discriminator (0.0223) losses. Complex variations within simple and compound characters along with various handwriting styles become captured by the model which makes it suitable for digital archiving applications and textscript recognition systems and cultural preservation efforts. The research shows that RDC-GAN represents an effective approach to create AI-generated Bengali textscript for minority languages.
Keywords: Bengali script generation, Regularized deep convolutional GAN (RDC-GAN), Syntheticscript generation, generative adversarial network, Kullback-Leibler divergence, Wasserstein distance
Received: 02 May 2025; Accepted: 31 Oct 2025.
Copyright: © 2025 Nath, Banerjee, Pal, Sarkar and Raval. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Chintal Raval, chintalraval.dit@charusat.ac.in
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.