Application of Single Image Super-Resolution in Human Ear Recognition Using Eigenvalues

Abstract

Ear recognition is a field in biometrics wherein images of the ears are used to identify individuals. Many techniques have been developed for ear recognition; however, most of the existing techniques have been tested on high-resolution images taken in a laboratory environment. This research examines the performance of Principal Component Analysis (PCA) based ear recognition in conjunction with super-resolution algorithms from low-resolution ear images. Ear images are first split into database and query images; the latter are first filtered and down-sampled, generating a set ear images of different low resolutions. The resulting low-resolution images are then enlarged to their original sizes using an assortment of neural network-based and statistical-based super-resolution methods. PCA is then applied to the images, generating their eigenvalues, which are used as features for matching. Experimental results on the images of a benchmark dataset show that the statistical-based super-resolution techniques, namely those that are wavelet-based, outperform other algorithms with respect to ear recognition accuracy.