Robust person authentication using dyadic wavelet filters learned by cosine-maximization

This paper produces a person authentication system using learned lifting dyadic wavelet filters. Our system has the capability of identifying persons whose facial image is changed the size and angle of an original image. The learning algorithm is done by training free parameters in the lifting filters so as to maximize the cosine between a vector whose components are the lifting filters and a vector of facial parts. This problem can be solved fast using Newton’s method. Applying the learned filters to a test image, facial parts in the image can be detected. Our detection of facial parts is robust for expansion, shrink and rotation of the image. A person is identified by checking the number of faces detected from video frames. Simulation results show that our person authentication is accurate and fast.