Evaluate and Determine the Most Appropriate Method to Identify Finger Vein

Abstract

In this paper, the performance of a variety of different methods of dimensionality reduction on finger vein database is evaluated to determine the most appropriate one in terms of finger vein recognition. Principal Component Analysis using K-nearest neighbor (KNN) as a classifier, different types of Kernel Principal Component Analysis (KPCA) using KNN as a classifier, different types of Kernel Entropy Component Analysis (KECA) using KNN as a classifier, and finally different types of KPCA using Local mean-based k-nearest centroid neighbor (LMKNCN) as a classifier are implemented on finger vein database. Different types of KPCA and KECA used in this experiment are Linear, Polynomial, and Gaussian. Extensive comparisons are made in this paper to identify which method matches finger vein recognition best.