The Biometric Recognition System Of Multispectrum Palm Features Based On Videos

With the continuous development of global economy, information security has become a severe problem. Among numerous solutions to this problem, one of the hotspots is to use biometric recognition system instead. As a member of biometric features, palmprint and palmvein seem to be with larger feature region, richer texture information and stronger robustness than the traditional fingerprint and fingervein features, thus more and more researches are focusing on this part. However, the existed study on palmprint and palmvein is still stay on analysising a single image, in which one the data acquisition process need to restrict the placement of palm to some extent for obtaining a clearer image. Based on the above method, the recognition system may become sensitive to noises like motion blur, defocus blur and local distortion of the palm, which will easily lead to an error result when identifying.Aiming at these problems, therefore, we proposed a new recognition system in this paper – the biometric recognition system of multispectrum palm features based on videos. By compared with the existed recognition systems, our contributions focused on the following parts:First, we proposed a new-fashioned recognition and registration method. In this way, a video about the palm moved above the acquisition platform will be collected instead of the single image, and then some ideal images will be filtrated for features extracting and matching. Thus in the register phase, this method can both enhance the robustness to gestures transformation and reduce the placement constraints of palm, consequently the stability and practicability of the recognition system will gain a great improvement. Besides, in the recognition phase, the users don't need to care about system operation, just waiting the feedback of system after swepting the palm through the acquisition platform, which makes the recognition system more affinity, convenience and interoperability.Second, we proposed a novel local invariant feature descriptor. By the effective combination of the double orientation histogram and the region division method based on intensity order, our descriptor performs excellent on recognition accuracy and speed. After experimental verification, our descriptor can obtain a better comprehensive performance than other common classical descriptors whether the palmprint images are clear or with slight motion blur. Therefore, applying this descriptor to our proposed recognition system will save a lot of time on the basis of increasing the recognition accuracy.Third, we proposed a cascaded pattern for score fusion. After extracting and matching the invariant features of palmprint, the ROI features of palmprint and palmvein, our system will fuse their scores with a cascaded pattern by utilizing their correlation. Comparing with the traditional weighted fusion, this cascaded pattern can improve the recognition speed of most registered users on the basis of reducing the equal error rate, which just accords with the actual demand of a recognition system.In addition, at the end of the experimental section, several groups of contrast experiments were conducted on the self-built database, which finally determined the optimal recognition algorithm for each feature and verified the advantage of our cascaded score fusion. During the examination of self-built database, the proposed recognition algorithm gained an EER of 3.97% in the case of most images were disturbed by the motion blur, which showed the effectiveness of our system.