Fingerprint Liveness Detection Using Curvelet Energy and Co-Occurrence Signatures

This paper proposes a new curvelet transform-based method to detect spoof fingerprint attacks in fingerprint biometric systems. It uses only one image to differentiate a real fingerprint from a spoof one. It is based on the observation that, real and spoof fingerprints exhibit different textural characteristics. Textural measures based on curvelet energy signatures and curvelet co-occurrence signatures are used to characterize fingerprint texture. Dimensionalities of the feature sets are reduced by running pudil\´s sequential forward floating selection (SFFS) algorithm. We test two feature sets independently on various classifiers like: AdaBoost. M1, support vector machine and k-nearest neighbor; then we fuse all the mentioned classifiers using the "majority voting rule" to form an Ensemble classifier. Classification rates achieved with these classifiers for energy signatures range from ~94.12% to ~97.41%. Likewise, classification rates for co-occurrence signatures range from ~94.35% to ~98.12%. Thus, the performance of a new liveness detection approach is very promising, as it needs only one fingerprint and no extra hardware to detect vitality.