Fonts Style Transfer using Conditional GAN

A font is an important element in designing printed materials including texts, such as documents, posters, leaflets, pamphlets, etc. Recently, many digital fonts with different styles are available for desktop publishing, but the number of Japanese-language fonts is smaller than that of European ones. This causes a problem when designing the materials including Japanese and European letters. Creating a new font is difficult and requires specialized knowledge and experience. Our research goal is to address this problem by transferring styles of the European fonts to Japanese characters by using a neural network. In this paper, we report some experimental results using the well-known deep learning framework called "pix2pix."