Presenting Text Layout and Color Patterns Suitable for Book Covers

Abstract

We propose a design support system that automatically generates book cover designs according to several parameters set by the user. Our method for book covers is based on the technology called "Neural Style Transfer" to generate the background image for the cover. Another neural-network-based framework called "Word2Vec" is used to select a font for rendering the book title. Several cover patterns are generated according to the user specified parameters. The user can perform post-process modifications on generated cover patterns to better match one’s taste. We conducted comparison experiments where participants were presented a pair of book covers, both of which represent the same book title, one generated by the proposed system, and another taken from an existing paperback in a store. The results of user studies showed that the existing cover design obtained higher evaluations, and that 80% of the participants gave high ratings to our method when evaluated alone. As our future work, we plan to apply our method to other design tasks, such as for cover belts for books (called "obi" in Japanese).