A low-rank LBP with local differential polarisation for fingerprint liveness detection

With the widespread use of fingerprint recognition technology in identity verification in recent years, fingerprint authentication system is at risk of being spoofed by fake fingerprints. Fingerprint liveness detection (FLD), a feasible strategy protecting fingerprint systems from presentation attacks, has become an academic hotspot. As a descriptor of local texture features, local binary pattern (LBP) is used to extract local texture features in a fingerprint image. However, features extracted by the LBP descriptor contain large quantities of noise, resulting in lower detection performance. In order to address the problem, this paper proposes a FLD method based on low-rank LBP (LLBP) with local differential polarisation (LDP). Firstly, the LBP descriptor extracts the feature matrix of the fingerprint image. Then, through robust principal component analysis (RPCA), the rank of the fingerprint feature matrix is reduced, eliminating negative effects of disturbing texture features. In addition, the blank areas of the image do not involve fingerprint-related information. The extracted features mixed with invalid information will interfere with the judgement of real and fake fingerprint detection and weaken the detection performance. Therefore, the LDP algorithm is proposed to remove the influence of blank areas. Finally, the features are fed into SVM classifier for subsequent model training and testing. Experiments, carried out on LivDet 2011 and 2013 datasets, show that our proposed method is superior to other methods.