The individual privacy has become increasingly important with the development of mobile internet. Compared to the traditional security methods such as tokens and passwords, biometrics are definitely more safe and reliable. They have been widely deployed for both public and private applications. Among the various kinds of biometrics, fingerprint recognition is the most popular technique due to its convenience, efficiency, stability and low-cost. Fingerprint recognition has been applied to commercial and governmental affairs extensively. However, there exists a crucial issue, the security of the fingerprint recognition system, which should to be taken seriously. In the daily life, fingerprints can be inadvertently left on the surface of objects that are touched or handled by a person. These fingerprints can be lifted by criminals and then used to produce fake fingerprints, which would threat the public and private securities gravely. In order to deal with fake fingerprints and enhance the security of existing fingerprint recognition systems, we propose several algorithms to distinguish fake fingerprint images from real ones, and explore the possibility of detecting fake fingerprints made of unknown materials. The contributions of this paper are summarized as follows. 1. We propose to detect fake fingerprints through sparse representation. As the fake fingerprint detection is essentially a special case of image classification issue, on which the sparse representation achieves outstanding performance, we attempt to recognize fake fingerprint using dictionary learning and sparse coding in this paper. LASSO is a classical model for dictionary learning and sparse coding. If there is a group of highly correlated variables, then LASSO tends to select zero or one variable from the group and ignore the others. In this paper, we introduce the elastic net model instead of the LASSO to train the dictionary. The elastic net model exhibits the grouping effect. The experimental results on the Fingerprint Liveness Detection Competition 2009 database prove that the proposed method outperforms most existing fingerprint liveness detection algorithms. 2. We propose the use of MSLBP (Multi-scale local binary pattern) for fake fingerprint detection. LBP is an efficient feature for texture classification, and it has been applied to fake fingerprint detection. However, the original LBP only operates in a very small area, and it cannot reflect the detail features in fingerpri...
修改评论