Evaluation of GMM approach to fingerprint classification
Anibal Cotrina, Jorge Leonid Aching Samatelo, Evandro Ottoni Teatini Salles · 2011
This paper investigates the modeling of the characteristic vector of a PCASYS approach to fingerprint classification problem. In a previous work, it is proposed a new algorithm based in multiple levels of representation to detect the reference points in a fingerprint. The results indicates that an unimodal Gaussian distribution models each of the fingerprint classes, in contrast of other results that indicates a Perceptron neural network as the best classifier. Therefore, in order to verify it, this paper suggests more accurate tests over the feature vector. Here, each class is supposed unknown and modeled by two approaches: GMM (Gaussian Mixture Model), classified by Normal classifier, and a Gaussian Mixture Based Classifier (GMBC). The tests are conducted using the DB4 database and the protocol suggested by the National Institute of Standards and Technology (NIST). Finally, the results are evaluated and discussed at the end of the paper.