Real-Time Facial Feature Detection for Person Identification System

In this paper, we present an approach to realtime facial feature detection. Facial regions are segmented using Gabor filter responses with M-style grid matching. M-style grid matching method has been shown more effective than Gabor bunch graph matching method in many aspects such as frontal face detection against expression, in-planeldepth rotation, and various illumination environments. In addition, this approach can be implemented with low computational complexity. The center positions of both eyes are detected, from the segmented face region, by iterative binary thresholding with perfect contour tracing. Comparing with other pattern matching methods, it is shown that our scheme is faster and more effective eye detection method. Offline simulation results using the test image set taken under office illumination (fluorescence) are over 99% successful segmentation rate of facial region (Face Detection Rate : FDR), and 99% effective eye center position detection rate of facial region (Eye position Detection Rate : EDR) We have implemented the real-time system on Pentium-III550MHz PC, and the system is capable of finding a pair of facial feature points on 240 by 320 images at 220ms per image. 97% of FDR, and 85% of EDR are real-time performance of the online system. The measured computational complexity is as low as about 32WMOPS.