Learning the Deep Features for Eye Detection in Uncontrolled Conditions

Although eye detection has been studied for a long time in academic and industrial communities, it is still a changeling problem if facial images are with varying head poses, facial expressions, illuminations and resolution changes etc., which tend to happen in uncontrolled conditions. In this work, we propose to learn deep features that could capture the appearance variations of eyes for eye detection on those changeling facial images. Specifically, we exploit the idea of deep feature learning method, and construct eye detector based on the learned features. Experimental results on benchmark databases with different head poses, expressions, illuminations or resolutions show the effectiveness of the eye detector based on the learned features compare to state-of-the-art works.