A novel matching algorithm for distorted fingerprints based on penalized quadratic model

At present, one of the most challenging problems in fingerprint recognition is the matching of distorted fingerprints. In this paper, we propose penalized quadratic model to deal with the non-linear distortion. Firstly, minutiae as well as sampling points on all the ridges are employed to represent fingerprint. Secondly, similarity between minutiae is estimated by their neighboring sampling points. Thirdly, greedy matching algorithm is adopted to establish the initial minutiae correspondences which are used to select landmarks to calculate the quadratic model parameters. At last, input fingerprint is warped and matching process is conducted again to obtain similarity score between warped fingerprint and template fingerprint. In order to diminish the impact of the erroneous landmarks, we introduce a penalty term into the quadratic model to keep it smoothing. Experimental results on FVC2004 DB1 approve that quadratic model is effective to describe the inner-image transformation of a quadratic skin surface, and the proposed strategy can improve the performance of fingerprint matching algorithm.