K. Matsuo, M. Hashimoto, and A. Koike (Japan)
Face recognition, subspace, principle components analysis,and lighting fluctuation
In this paper, we propose a dictionary registration method for subspace face recognition and realize face image recognition with strong tolerances to lighting fluctuations. Today, various useful functions have been added to cellular phones and camera devices are enjoying widespread popularity. The method has made biometric certification of face images possible without the addition of extra devices, such as a fingerprint sensor. However, face images taken by cellular phones are fluctuated by lighting conditions, because cellular phones are used both indoors and outdoors without environmental restrictions. Therefore, the most serious problem with this system is that lighting fluctuations decrease its recognition accuracy. Our proposed method is very simple. It uses the lighting canonical space which contains various types of lighting elements and was modeled in advance, and creates a virtual subspace that contains various types of lighting fluctuation elements without actually taking extra face images. Simulation results show that the proposed method can achieve not only a lower equal-error-rate of verification, but also higher precision of identification than the conventional subspace method.
Important Links:
Go Back
