Research Of Recognition And Authentication Algorithm Of Cancelable Facial Feature

Face recognition technology is increasingly mature, already obtained the widespread application. However, face data or face feature value is directly stored in the database in the traditional face recognition systems or authentication system based on facial feature,which has a great safety hidden trouble. Since biological characteristics is scarce and need to be protected, Ratha et al proposed a concept of cancelable biometrics. In recent years,the research reports on cancelable facial feature recognition technology are relatively small. Therefore, this paper focuses on recognition and authentication technologies of cancelable facial feature, the major contents are as following:(1) Random projection is extended to two-directional two-dimensional random projection. Thus the Biohashing scheme based on random projection theory is extended from one-dimensional to two-dimensional, come into being(2D)2Biohashing schemes, so that the storage and computational costs are both reduced. What is more, this paper discovers that binaryzation can be kept well in high dimensional feature in Biohashing scheme by experiment, and a improved quantization scheme is proposed in low dimensional facial feature.(2)(2D)2GL-Hash scheme is proposed, which combines(2D)2Biohashing with two-dimensional Gabor wavelet transform and(2D)2LDA. By use of the characteristics of multi information of Gabor wavelet transform and characteristics of good classification of(2D)2LDA, we can achieve the best recognition performance using appropriate dimension of facial feature. Then, with the good identification performance of binaryzation in high dimensional of feature, the scheme can achieve high security and high authentication accuracy.(3) Make full use of the characteristics of multi information of the Gabor wavelet,the scheme of multiscale scrambling fusion in feature level based on Gabor wavelet transform(Gabor-SF) has been proposed. First, data from all directions of each scale of Gabor features are encoded jointly in order to reduce the dimension of facial feature. Then scramble and fusion the data from all scale of Gabor features. Experimental results show that the scheme can achievee high security and high authentication accuracy in ORL andyale face database.In conclusion,the algorithm and scheme in this paper can effectively ensure the safety and privacy of the face feature,and achieve good accuracy of identification and authentication.