Adaptive SIFT-Based Algorithm for Specific Fingerprint Verification

The performance of an fingerprint authentication algorithm can be decreased significantly if the fingerprint image has lots of broken ridges caused by cutline, or the overlap area between the template and input is very small. For the purpose of these specific kinds of verification, a Scale Invariant Feature Transformation (SIFT) feature-based algorithm for fingerprint verification is presented. This approach is not based on traditional minutiae or ridge features. The SIFT keypoints in Gaussian scale-space and the local descriptor for each SIFT keypoint can be extracted by using this method. The verification is done by matching the descriptor, which is invariant to image scale and rotation. In this paper a proper pre-processing is carried out on the fingerprint image instead of using the original fingerprint image. This can make the algorithm adaptive to the variation of the impression condition. Furthermore, a Hough transform adapted to fingerprint verification is performed rather than only using SIFT keypoint descriptor matching. The fusion with minutiae information is also applied for efficiency and accuracy. Two specific databases are captured for experiments. Experiment results of proposed algorithm on specific databases show significant improvement compared with common minutiae-based method. Experiment results on FVC2002 Database show that Equal Error Rate (EER) and False Matching Rate (FMR) of our proposed algorithm can be decreased to about 20% of previous SIFT-based works.