Huang wins EAGER grant to improve search processes for multimedia information networks
In August, the National Science Foundation (NSF) awarded the University of Illinois at Urbana-Champaign, as well as the City University of New York, a one-year, $199,360 EAGER grant.
According to the NSF website, the goal of this project, titled, “Exploring Multimedia Information Networks,” is “to provide effective methods for organizing, searching, mining and reasoning with web-scale multimedia.”
Today, because of the ever-expanding amount of various kinds of information on the web, traditional multimedia search processes are facing challenges.
Through this project, researchers aim to overcome these challenges by creating a structured multimedia database, called Multimedia Information Networks (MINets), which will be able to link multimedia data by identifying semantic concepts in images, video and text.
However, “trying to derive semantic concepts from these is the most difficult part,” said Thomas Huang, who is the principle investigator (PI) of the project and a research professor at the Coordinated Science Laboratory. He is also a professor of electrical and computer engineering and a research affiliate in Beckman Institute.
“This is not a regular grant, it is an exploratory grant,” Huang said. This means that the project may or may not be successful, but the point of the research is simply to explore. “So there’s high risk, but potentially high return,” Huang said.
It is impossible to collect images by taking photos on your own and labeling them if you want billions of images for a database, Huang said. So in order to create this multimedia database, the researchers must crawl the web for already existing images, video and text.
Huang said that the researchers start by building a small database, such as a vehicle database that includes land, marine and air vehicles. Researchers collect multimedia data, such as photos of cars, for their database by crawling the web and can then use those photos to recognize other multimedia that wasn’t previously in their database.
One of the domains that researchers are currently looking at is natural disasters, Huang said. In order to create a database about natural disasters, they must first look at the key concepts of natural disasters. However, if they are creating a database of a disaster while it is currently taking place, then they have to constantly update the database.
“This kind of concept could be really useful if it succeeds, but it’s exploratory,” Huang said.
In addition to MINets being able to recognize semantic concepts in order to build its database, it is also expected to work properly in the presence of noise or uncertainty. Researchers will assess Quality of Information factors in MINets such as coherence, accuracy, recall, and freshness of information.
There are four main students who are working with Huang at the University on this project: Guo-Jun Qi, Min-Hsuan Tsai, Shen Fu Tsai and Shiyu Chang. All four are graduate students in electrical and computer engineering and their primary research area is in signal processing.
While Huang and his students use their background in signal processing, he said, they collaborate with Heng Ji, the Co-PI and an assistant professor at CUNY whose research interests focus on natural language processing.
“The most exciting thing is that it’s a fuzzy field … there’s many different directions you can explore,” Huang said. “You can invent new paradigms of doing things, like new ways of doing searches for multimedia data for example.”