Modelling and Analysing Orientation Fields of Fingerprints
Thomas K. Hotz · 2007
Global features of fingerprints, the most useful of which are orientation fields, are deployed for classification, image enhancement, and to define intrinsic coordinates. We thus strive for simple parametric models for orientation fields of fingerprints, geared towards these applications, whose parameters are invariant under Euclidean motions. Improving upon models from the literature by taking the special geometry of fingerprints into account, we propose models based on quadratic differentials. We develop algorithms to fit these models to given data whilst avoiding numerical instabilities caused by non-convexity of the functional measuring the quality of the fit. We then compare these models to models from the literature, demonstrating their superiority over the latter as far as accuracy and predictive performance are concerned. We finally analyse the models" potential for the applications described; they appear especially useful for extrapolation and for the definiton of intrinsic coordinates.