Research On Video Based Fingerprint Verification

Fingerprint recognition is the most wildly used identity authentication approach that based on biometric traits. There are many research achievements about automatic fingerprint recognition technology which has been widely applied. But the accuracy of Automatic Fingerprint Identification System (AFIS) is still much lower than accuracy acquired manually, which hinders its wider scale deployment. In recent years, a lot of organizations are devoted to improve the accuracy of AFIS to accelerate its wider deployment. Traditional fingerprint verification uses one single fingerprint image or several single images captured independently. In this thesis, we propose to use videos for fingerprint verification and explore the information contained in fingerprint videos for higher accuracy. There are three advantages to use video for fingerprint verification:(1) Fingerprint video contains more information than single impression; (2) user experience of providing a fingerprint video is the same as that of providing a single impression; (3) it is difficult to get the dynamic information contained in fingerprint video without normal capturing process.In this thesis, two video-based fingerprint verification methods are proposed based on the fingerprint video database built by our research group. The two methods explore and use the information contained in fingerprint video to improve the accuracy of traditional fingerprint verification methods. The proposed methods include:(1) Fingerprint video verification using "inside and outside similarity". We define "outside similarity" which measures the similarity between two matching fingerprint videos and "inside similarity" which measures the similarity between images inside a video. We calculate the final match score of two matching fingerprint videos using the two kinds of similarity. This method not only uses the static information in the single images, but also uses the correlation between images inside a video which is a kind of dynamic information of fingerprint video. Experimental results show that this method lead an improvement of 60% comparing to the traditional single image based method. And this method can improve the accuracy performance even with equal time complexity. (2) Fingerprint video verification using "Inclusion Ratio" and "Consistence of the Rotation and Translation Parameters". This method focuses on using the dynamic information in fingerprint videos. After preprocessing and aligning of a pair of matching fingerprint video, we acquire the matched minutia sets and the rotation and translation parameters. Then we calculate the inclusion relation between these minutia sets, i.e., the "Inclusion Ratio", and also calculate the consistency of these rotation and translation parameters. Finally, "Inclusion Ratio" and "Consistency of Rotation and Translation Parameters" are used to measure the match score between the matching fingerprint videos. Experimental results show that this method can improve the accuracy of fingerprint system. After fusion with single image based method, better accuracy can be accessed.This thesis propose a new idea that use videos for fingerprint verification, take full use of the abundant information in fingerprint video by one capture process, and provide a new approach for further improving the accuracy of fingerprint system. Studies show that video-based verification method can improve the accuracy of traditional single impression based method. However, as research on video-based fingerprint verification is fairly new, there is still much work to do. Future work will involve exploring different features contained in fingerprint video comparing with that in single impressions, and utilizing fingerprint video to alleviate some security issues.