Handwriting Image Generation by Adversarial Training

The generative adversarial network (GAN) has become a popular research direction in the field of artificial intelligence, which has been applied to the image and visual computing, speech and language processing, information security, chess games, etc. In this paper, using GAN to generate handwriting numerals, we propose a new framework to generate handwriting images through the confrontation process. To be specific, we train the generation model G and the discrimination model D iteratively. The generation model G aims to estimate the empirical data distribution. In contrast, the discrimination model D judges whether the input is a real data instance or a fake sample generated by G. The training process is an adversarial game. The generator learns to realistic synthetic data to fool the discriminator. In contrast, the discriminator tries to distinguish the fake data from the real data. In this way, G will gradually approach the empirical data distribution, thus deceiving the discriminator. As a result, the generator can generate plausible handwriting images. The experiment proved the potential of the framework by qualitatively and quantitatively evaluating the generated samples.