Goshuin 2.0: Construction of the World’s Largest Goshuin Dataset and Automatic Generation System of Goshuin with Neural Style Transfer
The goshuin is a vermilion stamped and inked text that can be obtained as a proof of visit to a shrine or temple. It has been in circulation mainly in Japan since the Middle Ages, and in recent years, the artistic quality of the goshuin has been appreciated and is attracting attention. However, it has received little attention in the academic field, and there are no examples of the goshuin research, especially in the field of information science. Furthermore, there are no data sets that can be used for research, and even on the Web, there are only about 1,000 copies of the goshuin at most. In addition, shrines and temples have a particularly low birthrate and an aging population, so the burden of writing the red seal is heavy. To solve the above problems, we construct a dataset system that allows users to view and download about 4,000 copies of the goshuin. Users can also narrow down their search by denomination, region, shrine or temple, set, etc., or by name. Furthermore, we build an automatic generation system based on Neural Style Transfer, which outputs images that have both various existing goshuin fonts and arbitrary text entered by the user. Through our system, creating goshuin in a variety of calligraphic styles without much effort. Further, we expect to increase the interest of people not only in Japan but also in the world, promote research on the goshuin, and make English-speaking people aware of the appeal of kanji and calligraphy.
Shuma Shimizu, Tatsuya Minagawa, James Harry Morris, Xanat Vargas Meza, and Yoichi Ochiai.