Periocular recognition under expression variation using Higher Order Spectral features

Abstract

Periocular recognition promises attractive trade-off between iris recognition and face recognition because it provides a longer range of imaging than iris and could achieve a higher recognition performance than face. This benefit is critical to the success of a biometric system under unconstrained and less cooperative operating conditions. A number of feature encoding techniques have been proposed for periocular recognition including global features (SIFT, HOG) and local features (LBP). However, these features perform poorly for images captured in non-ideal conditions with the presence of facial expressions. In this paper, we investigate periocular recognition dealing with the deformation caused by facial expressions. In addition, we also investigate a novel feature encoding technique, called Higher Order Spectral features, on periocular images. We show that our proposed approach toward features for the periocular recognition under facial expressions using Higher Order Spectra is effective in encoding discriminant features. The proposed approach is validated on the JAFFE dataset.