Ultibiometric Fusion Recognition Based On Face And Fingerprint Features

With the improvement of social security requirements in information society, traditional identification method is gradually replaced by biometric identification technology. With the characteristics of universal, high security and high accuracy, biometric identification technology make the identity recognition system safely, conveniently and efficiently. And avoids the drawbacks of traditional identification method which is easily lost, forgotten and/or vulnerable. As different single biometrics have different limitations during the application, multibiometric identification technology attracts people further attention and research. Multibiometric identification technology combines the characteristics of several biometric to improve the security and anti-attack capability of the recognition system. multibiometric system reduce the error rate, relieve the defects of single biometric identification technology, and achieve better recognition performance.There are four levels to fuse different biometric recognition:pixel level fusion, feature level fusion, score matching level fusion and decision level fusion. With excellent performance and the advantages in universal, the score matching level fusion becomes the main research topic in recent years. This paper makes an in-depth study to multimode biometric recognition by score matching level fusion.The paper implements face recognition algorithm based on PCA and fingerprint recognition algorithm. And creates a Dual-mode database using the public face and fingerprint databases by artificial pairing.Three score fusion methods are improved in this work:face-fingerprint recognition algorithm based on the improved GMM (based on density), method based on classification of trust weighted(based on score normalization), method based on improved SVM(based on classifier).By combining the merits of density estimation and score normalization methods, based on GMM(Guassian mixture model)) and WSUM(Weighted sums), the paper proposes a Multi-biometric two level fusion method. GMM model is used to build the probability distribution of scores and N-P criteria is adopted as the first-level fusion strategy; A weighted sums normalized fusion method is introduced as the second-level fusion strategy. The ORL and AR face database, FVC2004fingerprint database are used to build a face-fingerprint multimode database to evaluate the proposed method. Experiment results show that, the proposed method can effectively enhance the performance of recognition.