Finger Vein Recognition Using Principle Component Analysis and Adaptive k-Nearest Centroid Neighbor Classifier

Abstract

The k-nearest centroid neighbor kNCN classifier is one of the non-parametric classifiers which provide a powerful decision based on the geometrical surrounding neighborhood. Essentially, the main challenge in the kNCN is due to slow classification time that utilizing all training samples to find each nearest centroid neighbor. In this work, an adaptive k-nearest centroid neighbor (akNCN) is proposed as an improvement to the kNCN classifier. Two new rules are introduced to adaptively select the neighborhood size of the test sample. The neighborhood size for the test sample is changed through the following ways: 1) The neighborhood size, k will be adapted to j if the centroid distance of j-th nearest centroid neighbor is greater than the predefined boundary. 2) There is no need to look for further nearest centroid neighbors if the maximum number of samples of the same class is found among jth nearest centroid neighbor. Thus, the size of neighborhood is adaptively changed to j. Experimental results on theFinger Vein USM (FV-USM) image database demonstrate the promising results in which the classification time of the akNCN classifier is significantly reduced to 51.56% in comparison to the closest competitors, kNCN and limited-kNCN. It also outperforms its competitors by achieving the best reduction ratio of 12.92% while
maintaining the classification accuracy.