Facial feature detection in near-infrared images

We propose to employ near-infrared (NIR) images for face recognition in reduced illumination or total darkness. A homomorphic processing technique has been developed to effectively reduce the artifact of NIR images [1]. In this paper, we proceed to construct a facial feature detection system that would function independent of the surrounding lighting condition. Firstly, we propose a classification method based on local histogram analysis to separate NIR images captured at a short range from those in other circumstances. Afterwards, we present an algorithm to mark predominant facial features in a nearly frontal face NIR images acquired at a short range. Experimental results demonstrate that facial feature points can be located accurately in homomorphic-filtered NIR images.