Generative Adversarial Networks for Oracle Generation and Discrimitation.

Generative Adversarial Networks (GANs) have become the most prevalent generative models in recent years. However, due to the oracle image is a specially and rarely dataset, it brings huge challenge for interpreting and recognizing oracle information. Therefore, it becomes especially important to obtain more generated oracle images. In this paper, different from the original image, we use Gaussian pyramid (GP) to preprocess oracle, and analyze different generated models based on GANs. Experiment results show that our approach can synthesize highly realistic images of oracle bones, which is difficult to distinguish from the real oracles.