Research On Multi-modalbiometric Fusion Personalidentification Technology Based On Fingerprint And Name Speech

The recognition rate of the single-mode biometric identification will decline rapidly for the changes or deteriorations about identify environment. In this paper, the multi-modal biometric fusion personal identification which based on fingerprint and name speech has been proposed so as to improve the identification rate and adapt to poor recognition environment. this paper has discussed the normalized feature information firstly,and then the adaptive weighted fusion theory algorithm,D-S evidence theory fusion recognition algorithm and the fusion recognition algorithm of neural network theory. At last, experiments show that high accuracy, reliability and real-time are obtained even in poor identify environment to some extent.The main work and innovation of this paper as follows:1,A name speech feature extraction method which based on MB-FECC has been proposed. The speech has firstly been dealt with a set of bandPass Gammatone filters and improved by a triangular window. Then, the corresponding characteristic coefficient has been obtained by appling FrFT and ambiguity or instantaneous frequency function, and processed by DCT and logarithmic transformation so as to execute feature extraction. Finally, name speech personal identification has achieved by HTK.2,Fingerprint's feature extraction and identification has been achieved. First, fingerprint image has been enhanced by Gabor filter, and then refined twice through the sparse algorithm and eight neighborhood method which based on single pixel. Then, the fingerprint feature vector has been extracted by using two-dimensional discrete wavelet decomposition. Finally, fingerprint image has been decomposed two-dimensionally and discretely, and fingerprint identification has achieved by K-nearest neighbor recognition method with no rejection rate.3,Multi-modal biometric fusion identification model which based on fingerprint and name speech has been proposed. First, the integrated theoretical framework of fingerprint and name speech on the matching layer and feature layer has been discussed. Then, the adaptive weighted fusion theory algorithm and D-S evidence theory fusion recognition algorithm which both based on the matching layer, and the neural network theory's fusion recognition algorithm which based on the characteristic layer. Finally, the experimental result about three fusion methods has proved that multi-modal biometric fusion identification which based on fingerprint and name speech is more accurate and reliable than the single one.