Research Of Finger Vein And Finger Dorsal Knuckle Texture Based Biometric Identification

Safety and security of information have become the focus of public in the information age. Identification is the core technology in protecting personal information and security. Traditional identification technology can not prevent forgery or imposter due to its own disadvantages. However, the rising biometric identification has more stable characteristics, and is less likely to be forged. It has been adopted commonly by the development of computer technology and biometric sensing technology.Now the widely used biometric identification technology includes finger print identification, human face recognition, etc. The finger print capturing needs finger to be touched with special instruments. Thus, it is possible to get finger print of others and counterfeit it. Nevertheless, human face characteristics are not stable because of illumination, posture and so on. Researchers propose the second generation of biometric identification technology constituted mainly of palm and vein. It is more stable. The capturing is contact less. And the performance of anti-counterfeit is better.Based on that, we proposed a novel multimodal biometric identification framework by fusing finger vein and finger dorsal knuckle texture. Finger vein identification is based on live body, the characteristics maintain stable for a very long time. And it has perfect identification accuracy and good performance of anti-counterfeiting. Besides, finger dorsal knuckle contains plenty of texture. This texture can keep the same for a long time. The two characteristics have their own advantages, and they supplement each other. So we present fusing the independent two characteristics in the aim of improving the identification accuracy.In the thesis, we firstly establish a piece of efficient hardware facility. Then we give out traditional finger vein extraction method's disadvantages and provide our own modified finger vein extraction algorithm. After that, we compare several feature extraction methods which are widely used in biometric identification and select appropriate feature dimension reduction methods and design effective classifier. As a result, we construct a robust identification system with high accuracy. Furthermore, the thesis lists the fusion strategies of finger vein and dorsal knuckle texture on data level, feature level and decision level and provides results and analysis. Besides, we compare single identification with fusion methods and compare among the fusion methods. We also show the equal error rate performance of different identification methods. Our results have certain significance in multimodal biometric identification especially in local finger biometric identification.