An Iris Recognition Method Of Multiple Features Extraction And Fusion

Authentication based on iris features have many of the advantages of its uniqueness, universality, stability, non-intrusive, etc. So iris recognition is known as the most promising biometric technology. A complete iris recognition system consists of four basic parts: iris image acquisition, preprocessing iris image, feature extraction and encoding, and feature matching. In this thesis, the three sections which was preprocessing iris image, feature extraction and encoding, and feature matching as the main objects were researched. An iris recognition method which was used multiple algorithms to extract iris features and fused features in the matching section, was proposed and implemented by computer simulation. The main contents was as follows:(1) The development of biometric technology and the theory of iris recognition technology were introduced briefly, and the main content and significance of this thesis were presented. The composition of iris recognition system were described, and all aspects of the classical algorithms were analyzed.(2) In the section of iris image preprocessing, a method based on the rough and accurate location iris boundary was implemented. The morphological closing operation algorithm was used to eliminate the influence of the eyelash and light spots in the iris location section. This method of iris location reduced the time of the entire iris recognition methods, and achieved an ideal result of iris location(3) The limitations of single iris feature extraction methods was analyzed, and proposed Log-Gabor filters and Haar wavelet decomposition used to extract iris features respectively, and encoded two unrelated feature template. This method was not only extracted the iris local features but also extracted global features.(4) In the section of matching, two unrelated feature templates were used Hamming Distance and the Weighted Euclidean Distance to calculate and achieve two values of similarity. Then fused the two values of similarity by constructing the SVM classifier to matching and classification. Because the different feature extraction and encoding algorithm, iris feature templates had different database structures. This feature fusion method solved the problem of the different database structures. The effect of the proposed iris recognition method were proved by simulation experiment. The correct recognition rate and recognition time had improvements by comparing to other iris recognition methods.