Research And Implementation Of Face Liveness Detection Algorithm

With development of the biometric technology,face recognition had become the most common authentication technology in academia and industry,and it had been gradually applied to the commercial markets with development of deep learning.However,most face recognition systems have a great hidden risk which they were vulnerable to face spoofing attacks.Illegal users can attack the face recognition system by using face photos or videos of legitimate users.In order to address this problem,face liveness detection technology had been proposed by researchers.It can guarantee the security of face recognition system by using some ways to distinguish the live face or fake face.This paper will focus on the non-interactive ways to detect live faces which mainly extract features from images or videos to distinguish the live face or fake face.This paper analyzed the differences between the live faces and fake faces,then we proposed improved face liveness detection methods using the different features.this research work mainly contained the two aspects: firstly,according to the loss of image texture information produced by the face in re-acquisition,we proposed an improved face liveness detection method by using the Ms LBP feature and Spectrum feature.This method trained two weak classifiers by using two single feature,then learned the weight by using these two weak classifiers and combined these two features using the weights to train a strong classifier.Then,it used this strong classifier to distinguish the live face or fake face.The use of weights not only overcomes the shortcomings of the Ms LBP that only extracts local spatial features but also highlights the different importance of two features in combined features,it improves the performance of method to the greatest extent.Then,according to the different light reflection produced by the live face and fake face,we proposed another face liveness detection method by using the multi-direction color-gradient feature.This method analyzed the differences of gradient between the live face and fake face and proved the effectiveness of color gradient feature for face liveness detection.Then,it used the improved Roberts cross operator to extract the multi-direction color gradient feature of live faces and fake faces to complete the classification.The improved feature not only considers the color information of an image,but also further improves the gradient feature by considering the four direction to do weight fusion,it gains a better expression of the differences of the live face and fake face.Experimental results on Replay-attack and CASIA-FASD benchmarks showed that the proposed methods could effectively detect the live face,and compared with other state-of-the-art methods under the same evaluation standard,it showed that the proposed methods could effectively improve the performance of face liveness detection...