Biometric recognition robust to partial and poor quality fingerprints using distinctive region adaptive SIFT keypoint fusion

Quantity and the quality of detected discriminating features is very important in any recognition system. In minutia based fingerprint matching techniques, the performance depends upon the quality and quantity of the detected genuine minutia points. In this paper, we have demonstrated that using only minutia points for fingerprint matching does not give optimum results. We have proposed a feature level fusion technique using minutia and Scale Invariant Feature Transform (SIFT) keypoints. The extra step of reducing false matches is required when using the SIFT method. The proposed method adaptively selects the SIFT keypoints using the distinctive region criterion. This criterion ensures that only the SIFT keypoints which are at distinct regions from the minutia points are taken for the fusion process. The proposed fusion technique improves the recognition performance by 3.29% and 4.23% in terms of equal error rate (EER) on publicly available FVC2004 DB1A and CASIA dataset respectively. The experimental results also prove the efficacy of the proposed fusion technique in dealing with poor quality and partial fingerprints. A new partial fingerprint dataset is created by cropping the fingerprints in a FVC2004 DB1A dataset to evaluate the performance in the presence of partial fingerprints. The proposed technique reduces the equal error rate by 10.21% for the partial fingerprint dataset. The performance of the SIFT based fusion method is also compared with other multi-scale feature detectors such as BRISK, KAZE, AKAZE, and ORB.