Fingerprint Type Classification Using Learning Vector Quantization

Abstract

This paper proposes a fingerprint types classification algorithm using Learning Vector Quantization (LVQ) with FingerCode features. This algorithm assigns each fingerprint image to one of the five subclasses, according to the Henry system: Arch(A), Tented Arch(T), Left Loop(L), Right Loop(R), and Whorl Loop(W). The search for a specific fingerprint can therefore be performed only on specific subclasses containing a small portion of a large database, which will save enormous computational time. We use the feature vectors from FingerCode generation process to train with the LVQ classifiers. In our feature extraction process, the oriented components are extracted from a fingerprint image using a bank of Gabor filters, and a feature vector is computed for each oriented component. The feature vectors from the input image are classified using LVQ classifier. This algorithm has been tested the fingerprint database. For the 100 fingerprint images, the classification accuracy is 93 %, with 7 % error rate for 5-classes.