Deep Neural Networks for Korean Fonts Generation

Korean fonts consist of 11,172 letters, requiring tremendous time and effort to design a new font. This cause a lack of diversity in Korean fonts. Deep neural networks have recently brought significant improvements in various fields of computer vision. Herein, we investigate deep neural networks as a means of generating a new set of Korean fonts. Three deep neural networks have been utilized to transfer the style of one font to the other. These networks are trained with a pair of 2,350 letters and used to generate the remaining 8,822 letters of one font in a style of the other font. The results are quantitatively and qualitatively evaluated, suggesting that the deep neural networks could facilitate a cost-effective and rapid design of Korean fonts.