A New Approach in Fingerprint Matching based on a Competitive Learning Algorithm
|Author(s)||A.N. Ouzounoglou, P.A. Asvestas and G.K. Matsopoulos|
|Issue Date||September, 2012|
|Publishing Date||September, 2012|
|Keywords||Fingerprint image identification; minutiae point correspondence; self-organizing maps; similarity measure.|
A fingerprint identification algorithm based on a modification of the Competitive Learning Algorithm developed originally by T. Kohonen is presented. Given a pair of fingerprint images (template and input image) and a set of minutiae in the template image, the algorithm attempts to extract the corresponding points (if exist) in the input image. This is accomplished by determining the parameters of local transformations that match regions around the interest points of the template image to their corresponding regions in the input image. A properly defined similarity measure is used to conclude whether the two fingerprint images are from the same person or not. The advantage of the proposed fingerprint matching method, compared to corresponding previous similar methods, is based on the non-necessity of constructing minutiae points in the input fingerprint image. The proposed fingerprint matching algorithm was evaluated using the DB3 database of Finger Print Verification Competition (FVC2004). Image pairs of both known and unknown transformations were used for the testing. The overall performance for fingerprints of the same and different fingers was calculated in terms of Equal Error Rate (EER) and was equal to 0.0586.