Beckman Team Wins Awards at FERA Competition

A team from the research group of Beckman Institute faculty member Tom Huang collaborated with a team from the University of Missouri at Columbia to earn first and second place finishes at the recent Competition on Facial Expression Recognition (FERA).

Held in conjunction with the 2011 IEEE International Conference on Automatic Face and Gesture Recognition March 21-25 at Santa Barbara, Calif., FERA featured 13 teams from six countries participating in the Emotion Recognition competition. The joint team from Illinois and the University of Missouri ranked first in Persona-Dependent Emotion Recognition and second in overall (person-dependent as well as person-independent) in the Emotion Recognition competition.

Huang led the team, along with Professor Tony Han of UMC and his graduate students. Huang’s students on the team were led by graduate student Usman Tariq. Student team members from Huang’s group included Tariq, Kai-Hsiang Lin, Zhen Li, Zhaowen Wang, and Vuong Le, as well as postdoctoral researcher Xi Zhou.

Tariq said the team began working on the competition in December of 2010. They started with a FERA database that consisted of video recordings of 10 actors displaying a range of expressions, while uttering a meaningless phrase, or the term ‘aaah’. The competition required five discrete, mutually-exclusive emotion categories to be detected: anger, fear, joy, relief, and sadness, with the emotions to be labeled on a per video basis. There were 155 videos in the training partition of the competition, with seven subjects/actors and each emotion appearing 30 to 32 times.

The next step was the testing partition of the competition, which featured 134 videos with six subjects/actors. The teams were given access to the testing partition for six days before returning their results. There was a secondary test phase, in which the participants sent their code to the organizers or brought it to the conference for a live testing with a limit of four hours. The team took about two-and-a-half hours to do the live testing, which featured 50 videos.  

The Illinois-Missouri team showed a 100 percent performance for the subject dependent part of the competition, making it the only team to achieve this score, and were ranked No. 1 for this competition.

The team wrote a paper on their method, which was accepted for presentation in the conference. They have also been asked to submit a paper on their methodology to the IEEE Transactions on Systems, Man and Cybernetics – Part B (TSMC-B) Special Issue on The Facial Expression Recognition Challenge 2011.