Investigation of some statistical properties of fingerprint patterns

Bob Hastings

    Research output: ThesisDoctoral Thesis

    357 Downloads (Pure)

    Abstract

    [Truncated abstract] There has as yet been no definitive study on the statistical uniqueness of fingerprints. This has become an important issue following several well publicised instances of mistakes in identification by law enforcement agencies. The motivation behind this work was to examine the statistical behaviour of the features used in fingerprint identification. Given two fingerprint patterns that appear to match, indicating that they may originate from the same finger, the goal is to estimate the likelihood that such a match could in fact have occurred by chance between prints from di erent fingers. I first carried out a preliminary statistical analysis on a small database of ngerprint images consisting of matched pairs (one ten-print and one latent print from the same finger) in which the position and orientation of each feature had been manually extracted. A RANSAC methodology was employed to extract from any given pair of prints the spatial transformation that gave the best match in terms of the number of matching feature points, the candidate match points being initially selected on the basis of the configuration of other minutiae in the neighbourhood of the reference point. I demonstrate that such a matching procedure results in a significant number of false matches, indicating that the number of match points is not by itself a reliable means of identification. More detailed analysis of prints requires the automated extraction of such parameters as the ridge orientation field and the locations of the minutiae. Some improvements to existing methods of extracting this information were devised. An "oriented diffusion" process was developed to enhance the image by removing most of the tiny line breaks and irregularities in the ridges, which would otherwise be falsely identified as minutiae. I demonstrate that this offers some improvements over the common approach using contextual filters, which are sometimes effective but require a good a priori
    Original languageEnglish
    QualificationDoctor of Philosophy
    Publication statusUnpublished - 2010

    Fingerprint Dive into the research topics of 'Investigation of some statistical properties of fingerprint patterns'. Together they form a unique fingerprint.

  • Cite this