Face and speech recognition fusion method based on penalty coefficient and SVM

The quality of biometric sample acquired from different acquisition devices is higher, then the reliability of recognition is higher. For the same biometric sample, recognition method is better, then the reliability of recognition is higher. So this paper proposed a multi-biometric recognition algorithm using biometric sample quality and recognition expert reliability (PSVM for short). First, obtains the sample penalty coefficient and reliability penalty coefficients from the sample quality and the expert reliability, then deduces the overall penalty coefficient, finally, uses the overall penalty coefficient to modify SVM fusion recognition algorithm. The experiment uses the XM2VTS database, in this paper compares the HTER of PSVM, Bayesian, FLD, MLP, Mean methods and SVM, the experimental results show that the HTER of PSVM fusion algorithm is lower.