A GVF-based Model For Low Quality Fingerprint Orientation Extraction

The main features used in Automatic Fingerprint Identification Systems areorientation field, frequency field, minutiae and singular points. As one basic feature,orientation field can be used for singular points detection and the enhancement,classification and matching of fingerprints. So orientation field estimation is a veryimportant step in the progress of fingerprint identification. Many algorithms have beenproposed for orientation field estimation, such as gradient-based methods, model-basedmethods, filter-based methods and other methods. The gradient-based methods aresimple and precise, but susceptible to noise; The model-based methods are not sensitiveto noise, but have high computational complexity and performming bad in singular area.The filter-based methods are robust to noise but time-consuming. The traditionalalgorithms listed above have three serious drawbacks:1) The lower quality thefingerprints are, the bad performance it has;2) It cannot get good performance inbackground area;3) The model-based methods,such as Legendre Model and FOMFEModel, can cause sever error in singular areas.We proposed a GVF-based model for low quality fingerprints orientation extraction.The GVF Model has two significant advantages. First, it has a very lage influence area.Second, it can effectively suppress the noise because its iterative natrue. The algorithmconsists of two main steps:1) A quick smoothness diffusion to remove the noise and geta reliable orientation field in background area.2) Orientation field correction in singulararea. Besides this, in order to have a better performance, it is necessary to do the qualityestimation and creases detection. The computation of quality estimation is mainly basedon the response of the complex filter, while the creases detection is decided by theresponses of a family of filters with different directions. In order to verify theeffectiveness of our algorithm, we conducted matching experiments and test it on FOEplatform. The results shows that the GVF-based method can get good performance onboth good fingerprints and bad fingerprints.