Fingerprint Matching using Orientation and Enhanced Shape Context Descriptors

Abstract

The unique physiological characteristics along with high collectability and permanence attributed to fingerprint patterns have allowed fingerprint identification to be a highly effective tool in performing human identification and authentication. In recent times, Automatic Fingerprint Recognition Systems (AFRS) have encountered widespread applications in forensics, commercial/industrial systems, and consumer products. Along with high tolerance toward image quality and accurate feature detection, such systems have the additional requirement of robust and highly accurate matching algorithm(s) that deal with performance deteriorating aspects such as skin distortion, dirt, scaring, spurious minutiae, and near disjoint region overlap. Recently, the shape context descriptor was applied in [1] for minutiae-based fingerprint matching following the model of maximal correspondence. This was achieved first by an alignment process using a novel greedy algorithm and Thin Plate Spline (T.P.S) transformation, followed by the computation of a matching score. Careful analysis revealed that minutiae-based matching algorithms like [1] strongly depend on the accuracy of minutiae extraction. Spurious and missing minutiae can both affect the result of minutiae correspondence, thereby degrading the overall matching accuracy. In this paper, two new variants of a robust fingerprint matching algorithm by integrating the shape context and a local ridge orientation descriptor are proposed. Experimental results on FVC2002 and University of Bologna databases confirm that the new algorithm achieves a significant improvement in overall matching accuracy, out performing many other well known algorithms despite minimal attention toward spurious minutiae removal.