2/13/2013 Kim Gudeman
Written by Kim Gudeman
Coordinated Science Laboratory resident professor Thomas Huang, along with his team of students, received one of two 2007 DoCoMo USA Labs Innovative Paper Awards at the IEEE International conference on Image Processing this fall.
The Illinois paper, co-authored by students Yun Fu and Xi Zhou, beat out more than 840 other submissions that were accepted by IEEE.
The team’s research unveiled a framework that enhances machine learning and pattern recognition. The framework enables the creation of new algorithms that solve semi-supervised multi-class object classification problems, making it easier and faster for machines to classify images.
The research focuses on semi-supervised learning, or using labeled and unlabeled training samples to help design classifiers. Fu and Zhou programmed about 20 objects, including toys, cars, etc., into the system. They took dozens of images of each object from different angles so that the computer would be able to recognize the object no matter what the view. The software is so advanced that the machine can recognize an object even if presented with few labeled training images.
The technology has a wide range of applications. It could enhance search engines by bringing up more accurate results when a user wants to search images. The program could also provide new methods in security applications, making it easier to detect a person from video sequence in an online-learning scenario.
While Fu and Zhou have already broken ground with their research, they will continue working to gain a better understanding of the science behind the technology.
“We know it works well in practicality, and we will further exploit the basic mathematical foundations to support the results,” Fu said.
Fu and Zhou are students and Huang is a professor in the electrical and computer engineering department. Zhu Li, a researcher at Motorola Labs, also contributed to the research and paper.
The award was presented in San Antonio in September.