Ear Biometrics by Force Field Convergence

Abstract

We have previously described how force field feature extraction can be used to exploit the directional properties of a force field generated from an ear image to automatically locate potential wells and channels which then form the basis of characteristic ear features. We now show how an analysis of the mechanism of this algorithmic field line approach leads to an additional closed analytical description based on the divergence of force direction revealing even more information in the form of anti-wells and anti-channels. In addition to furnishing specific implementation details for much faster FFT based computation and demonstrating brightness insensitivity, the technique is validated by achieving a recognition rate of 99.2% on a set of 252 ear images taken from the XM2VTS face database. These results demonstrate the inherent automatic extraction advantage of the new technique, especially when compared with more traditional PCA where we show that the ear set has to be more accurately extracted and registered in order to achieve comparable results. We show that it performs even more favourably against PCA under variable brightness conditions, and we also demonstrate its excellent noise performance by showing that noise has little effect on recognition results. Thus we have introduced a powerful new extension to complement our existing technique and we have validated it by achieving good ear recognition results, and in the process we have contributed to the mounting evidence that the human ear has considerable biometric value.