ASCII Art Synthesis with Convolutional Networks (final copy)

ASCII art is a type of graphic art that presents a picture with printable characters. It is commonly used for graphical presentations in text-based media. ASCII art can be categorized into two major styles: one is tone based, and the other is structure based. Tone-based ASCII art represents the intensity distribution of the original images by using the density of the characters. In contrast, structure-based ASCII art represents line structures of the original images using the directions of the lines in the characters. Thus, it is generally accepted that creating structure-based ASCII art is more demanding and difficult than creating tone-based ASCII art. There have been several attempts to automatically synthesize structure-based ASCII art [1–3], but there remains room for improvement in the quality of their products compared to ASCII art created manually by artists. To improve the quality of automatic synthesis, we develop a convolutional network (CNN) model [4] by training the network with ASCII art created manually by artists.