Key points are not available for this paper at this time.
There are many ways to cheat Biometric facial recognition systems such as recorded movies or portrait photographs.Hence, these systems need Liveness detection in order to guard against such attacks. We have proposed a new real time and single image Liveness detection and face identification approach utilizing Zernike moments and Fresnel Transformation. The advantages of using Fresnel transformation and Zernike moments to express the facial features are investigated in both face identification and Liveness detection scopes. A publically available PRINT-ATTACK database is used for evaluation of our Liveness detection method. Some of the conventional Liveness detection systems use 3D or IR cameras that are costly and may decrease the facial features that are important in face recognition. Multimodal biometric systems use several independent biometrics, like face and voice, simultaneously. Such methods need extra equipment and algorithms that may be expensive and time-consuming. Thanks to the ability of digital Fresnel transformation and Zernike moments to describe and differentiate the light intensity reflections and the aliasing characteristics, a common digital camera is used instead of 3D or IR cameras. The Fresnel transformation of the facial images is extracted and the Zernike moments are then calculated as the features for both face recognition and Liveness detection. A support vector machine classifier is used for Liveness detection and the hamming distance between the extracted feature vectors and the average of registered samples are calculated for face recognition. We obtained an accuracy of 94.0% in separation of the original face pictures and fake ones and 97.16% in face identification. Our methodology proposed a new generative rotation and scale invariant facial anti-spoofing approach that can be used instead of the state of the art features like LBP and Gabor wavelets. Keywords: Face Liveness Detection, Fresnel Transformation, Zernike Moments
Mousavizadeh et al. (Sun,) studied this question.