Some Unusual Experiments with PCA-Based Palmprint and Face Recognition

摘要

In this paper we describe a number of experiments relating to PCA-based palmprint and face recognition. The experiments were designed to determine the influence of different training sets used for the construction of the eigenpalm and eigenface spaces on the recognition efficiency of biometric systems. The results of the recognition experiments, obtained using three palmprint databases (PolyU, FER1, FER2) and one face database (XM2VTSDB), suggest that it is possible to design a biometric recognition system that is robust enough to successfully recognize palmprints (or faces) even in cases when the eigen-spaces are constructed from completely independent sets of palmprints or face images. Furthermore, the experiments show that for PCA-based face-recognition systems with an eigenspace that is constructed by using palmprint-image databases, and PCA-based palmprint-recognition systems with an eigenspace that is constructed using a face-image database, the recognition rates are unexpectedly improved compared to the classic approach.