The project was started to explore the importance of visual images for musical artists. The hypothesis was that the promotional photographs of artists encode some visual information about the style of music of that artist. We chose to obtain a large set of images of artists from Last.fm and we then proceeded to use a powerful image annotation system proposed by Makadia et al.
After seeing that automatic image annotation with genre tags performs better than would be expected if this process were random, we proceeded with album covers as well as a user study. This web site works to illustrate the workings of the automatic system.
This research has been published as "Exploring 'Artist Image' Using Content-Based Analysis of Promotional Photos" presented at the International Computer Music Conference 2010 in New York on June 2nd. The paper can be accessed here (pdf) and the presentation can be downloaded here (pdf).