Brian Kenji Iwana and Seiichi Uchida at Kyushu University in Japan have employed a deep learning network to study book covers and determine the category of book they come from. This will eventually set the stage for AI to design book covers without the need for human input.
The study took 137,788 unique book covers from Amazon and trained a neural network to recognize the genre by looking at the cover image. The algorithm listed the correct genre in its top 3 choices over 40% of the time and found the exact genre more than 20% of the time. “This shows that classification of book cover designs is possible, although a very difficult task,” say Iwana and Uchida.
The network had a very easy time in determining the genre of science-fiction, fantasy, travel and books about computers. This is because artists and graphic designers tend to use similar images for all of these genres. In retrospect it did struggle with books with pictures of a specific person, such as an autobiography or chef.
I think it is only a matter of time before this algorithm is refined enough that it can correctly identify the genre of book 99% of the time. This will set the stage for AI to be able to design book covers on their own, by taking into account hundreds of thousands of similar books.
via Cornell University