The Research Of Na?ve Bayes Classification Based Fake Fingerprint Detection

With the development of biometric information acquisition, as well as the innovation of feature extraction and matching, biometric identification technology is now in widespread use. Fingerprint is one of most well-known and widely used biometric. Fake fingerprint detection is a way to identify whether a fingerprint is from a live finger or not. This paper mainly focus on software-based fake fingerprint detection method, for the reason that it has lower cost and user intrusion, and can be integrated into existing equipment. The features extracted from fingerprint images are used to detect liveness, in other words, it is a pattern recognition problem.Though fingerprint can be forged, the materials used to make fake fingerprint are different from real skin. And the differences can be displayed in the extracted features. In order to enlarge the fingerprint database and extend the applicable scope of fake fingerprint detection method, fingerprint simulation and rolled fingerprint construction are researched in this paper. This paper proposed a wavelet analysis and local binary pattern(LBP) based fake fingerprint detection method. After image preprocessing, wavelet analysis is applied to get de-noised image and noise image. LBP features are extracted from these images and be selected by feature selection method. Finally, Support Vector Machine(SVM) is used to train and classify. Spatial surface coarseness analysis based method gets the noise image first, then extracts the standard deviation image and calculates the histogram. The histogram is regarded as the features to distinguish fake fingerprint from real one. For single classification may cause inaccurate result, this paper applied na?ve bayes classification method. The features extracted by local binary pattern method, space surface coarse analysis and curvelet transform are used to get classification results, then na?ve bayes classification can be utilized for making the final decision. The algorithm has been tested on the datasets of the second edition of the Fingerprint Liveness Detection Competition(LivDet2011) and the third edition of the Fingerprint Liveness Detection Competition(LivDet2013). The results compared with the winners show that this method has better classification effect.