Contourlet Based Image Compression for Wireless Communication in Face Recognition System

This paper proposes to use Contourlet transform for image compression and feature extraction for wireless face recognition system. The properties of face images and face recognition techniques are incorporated into the design of wireless transmission for such a system. The reasons for utilizing contourlet transform are two-folded. Firstly, in face recognition, the edge information is crucial in deriving features, and the edges within a face image are not just horizontal or vertical. When the coefficients are transmitted through the fading channel, the reconstruction from the Stein-thresholded noisy coefficients by contourlet achieves less mean square error than by wavelet. Secondly, when the network resources limit the transmission of full-set coefficients, the lower band coefficients can serve as a scaled-down version of the face image, for a coarser face recognition as screening. A prioritized transmission of the coefficients take full advantage of the wireless channel. Simulation shows that the wireless face recognition system works as well as a wired one, while gaining the cost efficiency, and the flexibility in deployment. An interesting phenomenon is discovered on FERET database that when the transmission error rate is increased linearly, the recognition performance degradation is not linear; instead, the performance stays the same for a large range of error rates, which illustrates that contourlet based face recognition system can tolerate the transmission error up to some threshold.