Synthetic Ancient Tamil Character Generation Using GAN

Stone Inscriptions are ancient handwritten scripts that comprise heritage information and are engraved on the stone walls of ancient temples and other structures. To conserve this priceless ancient information, it is crucial to digitize and decipher the information in these inscriptions. Due to the limited datasets that are available for these characters, recognition of a large number of characters is a difficult undertaking. Since ancient and modern Tamil characters are different from one another, it is not possible to utilize other Tamil character datasets for this purpose. Other data augmentation approaches won’t be effective in this situation since the structure of each character must also be preserved because there is very little variation between characters in old Tamil scripts. Using generative adversarial networks, the aforementioned problem can be effectively solved. GANs, or Generative Adversarial Networks, are a method of generative modelling that employs deep learning methods such as convolutional neural networks. In order to boost the effectiveness of character recognition, a method for the development of datasets for ancient Tamil characters using GAN is proposed here. According to a thorough simulation, character recognition from ancient stone inscriptions is more accurate when using an augmented dataset with GAN.