Liveness detection using frequency entropy of image sequences

Spoofing attack (or copy attack) is a fatal threat for biometric authentication systems. In this paper, we present a novel liveness detection method based on frequency entropy of image sequences against the photograph spoofing. By splitting the color video of face region into RGB channels, we can obtain the time sequences of each color channel. Then, three RGB sequences are analyzed through Independent Component Analysis (ICA) algorithm to eliminate the cross-channel image noises. Moreover, the Fast Fourier Transform (FFT) is further applied to these signals to get the power spectra of each RGB channels. Finally, the power spectra are verified through the entropy calculation to validate the liveness or photo attack. The experimental results demonstrate the superiority of the proposed method, which has an accuracy of more than 95%.