The Research Of Key Technologies Based On Multi-model Biometric Feature Fusion

In modern society,with the rapid development of network and information technology,the security of information becomes more and more significant.The biometric authentication technology is replacing the traditional identification technology such as password and key gradually because of its conveniences,safety and reliability.However,most of biometric authentication is based on uni-biometric systems.In practical application,uni-biometric usually have high rate of unacceptable error rates due to a series of problems such as the sensor noise,spoof attack,non-universality.Some of limitations mentioned above can be handled by multi-biometric systems.Therefore multi-biometric systems are supposed to have a good development prospect by researchers because of its high security and reliability.Face is the most widely used biometric feature,fingerprint have high identification and finger vein has high security.Therefore,it is of great value and great significance to research the system of multi-model biometrics base on face,fingerprint and finger vein.The main innovation points of this paper are list as follows.(1)A fusion method based on extract the feature vector for twice at feature extraction level is proposed.First extract the traditional feature of face,fingerprint and finger vein respectively.The parameters of GMM which are mean,variance and weight are utilized present the three different features.Then the method of feature vector is used to compute the partial derivatives of the mean,variance and weight of original features.In order to further explore the connection of multi-modal features,the fisher vector is extract secondly,in other words,the three fisher vector of three features will be cascade and become a new feature,then calculate the fisher vector of the new feature and load it into classifier to identify.(2)Based on the method of feature fusion proposed above,the paper studies the fake feature which appears in the practical scene.The liveness detection is append to the system,detect the picture is real or fake based on DCT,then remove the fake picture to reduce the influence of accuracy rate,and increase the robust of system.