Research On Algorithm Of Hand Vein Recognition

With the rapid development of information technology, personal authentication is becoming more necessary and even crucial in many fields. However, the traditional token or knowledge based personal authentication methods cannot meet the current requirements of society due to their inherent defects. Under such circumstances, biometrics emerged accordingly.As an important member of biometric family, vein patterns rely on the interior biological information of the body, and therefore, cannot be easily damaged, changed or falsified. Hence vein recognition is becoming one of the most reliable members of biometric family and has attracted extensive interests from the biometric researchers.Traditional vein recognition systems employ single vein trait, while in many aspects including accuracy, noise resistance, universality, multimodal biometrics systems perform better than those based on single feature. At that point, a new hand vein recognition system, which extracts and combines the dorsal, palm and finger vein for personal recognition is proposed.This thesis has done a wide range of studies in related work on hand vein recognition and survey on most possible algorithms. Main contributions are as follows:Firstly, comparison experiments are made on vein segmentation. Among the most common method for segmentation, the multi-scale 2-D Gaussian filter method is selected for the reason that it can not only get a good binary result, but also can be used to enhance the vein patterns. The enhanced images are used for the following vein feature extraction.Secondly, a new vein feature extraction method is proposed. Based on the enhanced images, Different Filter Bank (DFB) is applied to get the multi-directional subbands. Then the orientation LBP operator, which using 3-bits to represent the orientation of the basic LBP feature, is calculated to extract vein feature for each subband. Experiments show a well-proof result on higher effectiveness of the proposed method.Finally, a fusion method of multi-vein patterns of hand is also investigated. In this thesis, using SVM as the classifier, dorsal, palm and finger vein are fused at the decision level. This makes the system more robust since the recognition result not only depends on a single vein trait any more.The proposed method is tested on a large database (197hands, 3919 vein images) and the result is convincing with equal error rate (EER) of 0.0351% to prove the effectiveness of the method.