Eye Detection in Facial Images with Unconstrained Background

This paper presents an efficient eye detection approach for still, grey-level images with unconstrained background. The structure of the eye region is used as a robust cue to find eye pair candidates in the entire image. Eye pairs are located by a support vector machine-based eye verifier. The eye variance filter is then used to detect two eyes in the eye region which has been extracted in the eye pair location step. The proposed method is robust against clustered background, moderate rotations, glasses wearing, and partial face occlusions. The method is evaluated using the BioID face database. The experimental results demonstrate the effectiveness of the presented method.