Research On Face Liveness Detection

Network development and network security bring about more attention tobiometric based on-line authentication. However, on-line authentication proposes anew problem of biometric spoofing. Liveness detection is one of new researches inbiometrics. In this thesis, we focus on face liveness detection, the details are asfollows:First, we improve a liveness detection scheme to combining Fourier spectrumstatistics and local binary patteren (LBM). The image is preprocessed using Gammacorrection and DoG filtering. The illumination variation is reduced and the keyinformation in the image is preserved. Then the statiatical features of fourier spectrumare extracted. These features and LBP are combined together for classification. Theexperimental results on the NUAA demonstrate that the proposed scheme is efficientand robust.Second, we propose a face liveness scheme to combining gray levelco-occurrence matrix (GLCM) and wavelet analysis. We first compute the gray levelco-occurrence matrix (GLCM) of the image. Then we extract four features: energy,entropy, moment of inertia and the correlation on the basis of GLCM. In addition, wedecompose the image using Haar wavelet. And the mean values and energy of the firstand second high frequency subband are computed respectively. The experimentalresults confirm the efficiency of the proposed algorithm.We consider the live face detection as a binary classification problems. Weverify the efficiency of local binary pattern on living face detection. And we give thecompared experimental results.