Comparison of neural network based fingerprint classification techniques

The primary task of this work is to compare classification techniques to decrease the matching time in fingerprint identification. For classification, four different artificial neural networks are tested, as well as a non-linear support vector machine. All classifiers are compared and discussed to find the most suitable one. Automatic fingerprint identification systems (AFIS) are today widely used, but for use in embedded systems with less computational power, it is necessary to create less time-consuming systems. The classifiers splits a fingerprint database into four different subclasses. A multi-layer perceptron network using a backpropagation algorithm has shown to suit this problem best, outperforming both BAM, Hopfleld, Kohonen and just barely SVM, with a correct classification rate of 88.8%. This classification decreases the average matching time with a factor of 3.7.