Unsupervised Typography Transfer

Typography plays an important role in publishing industry, and typically different typographies are needed for different use cases. However, designing a new font is very expensive and time-consuming, especially for Chinese, which involves a huge amount of characters. Some semi-automatic typography synthesis approaches have been proposed. These methods utilize structural commonalities to simplify this design process. One approach [1] is to design a subset of characters for new fonts manually first, then generate remaining characters automatically. In this project, we will try to explore this problem using Generative Adversarial Network (GAN).