Doug Turnbull, assistant professor of computer science, combined his love of technology with his passion for music to prove that, with the right tools, an album can indeed be judged by its cover.
Turnbull recently published “You Can Judge an Artist by an Album Cover: Using Images for Music Annotation,” a research paper about how album cover art and promotional photographs can help categorize an artist into a specific musical genre.
Staff Writer Chloe Wilson spoke with Turnbull about his work with music and technology, the benefits of undergraduate research and the evolution of the media industry.
Chloe Wilson: How did you conduct your research?
Doug Turnbull: What we do is we look at images, both promotional photos and album cover artwork, and we analyze both the color content and the textures. There’s some information there that we’ve used to say something about the artists, and it works for some genres better than others. It’s not as good at analyzing the actual audio and humans, but there’s information in album covers and promotional photos, and we’re using a computer to prove that.
CW: Do different musical genres have specific image traits that help identify them from other types of songs?
DT: For pop music, you can imagine, they’re airbrushed — so very soft edges. But for techno music, maybe some covers would have lots of complex computer generated graphics, which have sharp edges. So we look at texture and we learn from them. We look at a couple hundred images associated with rock or hip-hop or blues, and we see if there are common colors and textures in those patterns. Then we take an image we’ve never seen before and see which one it is closest to in order to classify the image by the music genre
by comparing the colors and textures of this new image to the ones we studied.
CW: Do you think the importance of imagery has been influenced by the evolution of the media industry during the past decade?
DT: Yes, definitely. Album cover artwork has always been important because you have these big, beautiful album covers. And then we went to cassettes and CDs, which are smaller. Now we watch things like Pandora, or Last.fm, and they are full of images. So the album cover, being the iconic symbol of a particular song or set of songs, has lost that a little bit. We have so much imaging — you can go to any website and see thumbnail pictures of all these artists. We live in a multimedia world, and sounds and images and videos are a major part of that experience. Media, image and music all are related.
CW: Did you create the computer system used to analyze the images from music yourself?
DT: It was actually an undergraduate student project. I was a visiting professor at Swarthmore, and it was a student project. I teach a class on computer perception, which is analyzing images and sounds using computers. This was his project and then it turned into independent work and then he got good results — better than we expected — so we wrote up the paper.
CW: Do you encourage your students to complete undergraduate research projects?
DT: Students do these great projects, and sometimes they find
interesting results. [This project] is a good application of computer vision, which is a hot research topic, and people go to grad school for it. My student is now doing a Ph.D. in grad school on computer vision, and I study everything music, so this is sort of a merger of those two ideas.
CW: Are you continuing this type of work?
DT: I’m working on a personalized Internet radio system, which is like Pandora, and the interesting and new component of this is that it focuses on local music recommendation in particular, recommending music in the local Ithaca community. It’s like Pandora meets Craigslist. It’s a little more complex than your typical Pandora system, but we’re hoping that people who are really interested in learning about music, and if they’re new to a community like a new college student, they might find the system helpful.
To read “You Can Judge an Artist by an Album Cover: Using Images for Music Annotation,” visit www.jimi.ithaca.edu/~dturnbull.
