Skeleton-based Generative Adversarial Networks for Font Shape Style Transfer: Learning text style from some characters and transferring the style to any unseen characters

Abstract

This paper presents a new font shape style transfer technique that employs a generative adversarial network (GAN) and skeleton-based input feature maps to modify a target text to match a target font shape while retaining the original text content. Our GAN model is modified from a Shape-Matching GAN which utilizes a StyleNet generator and a PatchGAN discriminator. Rather than using a base-font character images as input to the generator like other existing font transfer models, we utilize the proposed skeleton-based features as input. The experimental results show that our model can produce the unseen characters in the desired font style better than an existing method.