June 16, 2009
Yung-hui Li, received the Best Student Paper Award in International Conference on Acoustics, Speech and Signal processing 2009 (ICASSP) held in Taiwan this spring. Li's advisor is Marios Savvides, assistant research professor of ECE. Li is a graduate student in the Language Technologies Institute.
Their paper, titled "A Pixel-Wise, Learning-Based Approach For Occlusion Estimation of Iris Images in a Polar Domain," proposes a novel way to solve the problem of iris mask estimation which has proven to be both efficient and accurate compared to existing methods in the literature.
In iris recognition, before researchers can employ pattern recognition, they need to perform iris segmentation and iris mask estimation. The goal of iris segmentation is to localize the eye in a given image. The goal of iris mask estimation is to estimate which regions are true iris texture and which are not. Once the true iris region is determined, researchers can then perform pattern recognition. If the iris masks are not estimated correctly, no matter how good the feature extraction and matching algorithms are, the iris recognition performance would be seriously degraded.
In this work, the authors treat the problem of iris mask generation as a machine learning problem, which is a completely new perspective in this field. By employing a Gaussian Mixture Model learning technique on local image features, they are able to successfully create an iris mask which is accurate (compared to manually created mask) and also very efficient (compared to other existing methods). Experimental results also show the created mask substantially enhances the iris recognition performance.
Graduate student Yung-hui Li
Assistant Research Professor Marios Savvides