A Robust Model based Algorithm for Detection of Singularities in Fingerprint Images

The performance of fingerprint recognition is heavily depending on the reliable extraction of singularities. Common algorithms are based on a Poincáre-Index estimation which is a numerical method. These algorithms ignore the topology of the underlying data and are only robust when certain heuristics and rules are applied. In this paper we present a model based approach for the detection of singular points. The presented method exploits the geometric nature of linear differential equation systems. Our method is robust against noise in the input image and is able to detect singularities even if they are partly occluded. The algorithm proceeds by fitting parameters for a given patch and then analyses these parameters. The parameters give a measure for the significance of the underlying structure and the type of a possible singularity. Corollary, our algorithm also reconstructs the direction field near a singular point which greatly aids the extraction of minutiae in these regions.