Eye detection using eye filter and minimisation of NMF-based reconstruction error in facial image

Abstract

A novel approach for eye detection is proposed, where an eye filter considering textural characteristics of eye regions and non-negative matrix factorisation (NMF)-based image reconstruction is utilised. The eye filter for detecting pairs of eye candidate points is utilised. To detect the exact eye position, all pairs of the eye candidates are evaluated by a cost function, which minimises the NMF-based image reconstruction error. Experimental results have shown that the proposed method achieves better performance of eye detection than existing methods under various environments such as head pose, facial expression and illumination conditions.