Autocorrelation and DCT based quality metrics for fingerprint samples generated by smartphones

It is becoming feasible and promising to use general purposed smartphone cameras as fingerprint scanners due to the rapidly improvement of smartphone hardware performance. We propose an approach to qualify the fingerprint samples generated by smartphones' cameras under real-life scenarios. Firstly, our approach extracts 6 quality features for each image block divided from a fingerprint sample using ridge patterns' spatial autocorrelation in both the spatial and the discrete cosine transform (DCT) domain. Secondly, a trained support vector machine is adopted to generate a binary decision to indicate the quality of the image block. Finally, we take the normalized count of qualified blocks as an indicator of the whole fingerprint sample's quality. Our experiments demonstrate that the proposed approach is effective to assess the quality of fingerprint samples captured by such general purposed smartphone cameras. A Spearman's rank correlation coefficient (ranging between [-1,1]) of 0.6354 is achieved between the proposed quality metric and samples' normalized comparison scores (as a ground truth) in our experiment.