In this paper, the problem of eye detection in the Middle-Wave Infrared (MWIR) spectrum is studied in order to demonstrate the importance of performing eye detection in the thermal band. While currently there are methods that are capable of performing automatic eye detection efficiently in the visible and active infrared (IR) spectrum (i.e., Near-IR and Short-Wave IR), eye detection in the thermal band is a very challenging problem. This is because in the thermal domain limited features can be extracted from the eye region, mainly eyelashes and eyebrows, while features such as human irises, pupils, and superficial blood vessels of the conjunctiva are not clear. Our proposed eye detection method operates in the MWIR band by combining a set of methodological steps such as face normalization, integral projections, and template-based matching. In this paper, a face database in the MWIR spectrum of 50 subjects is first assembled and used to illustrate the challenges associated with the problem. Then, a set of experiments is performed in order to demonstrate the possibility for eye detection in the MWIR band. Experiments show that (i) human eyes on still frontal face images captured in the MWIR wavelength band can be detected with promising results, (ii) that MWIR face images can efficiently be matched to MWIR face images (same session) using both research and commercial software (originally not designed to address such a specific problem), and (iii) the problem of matching MWIR images from different sessions is challenging. To the best of our knowledge this is the first time in the open literature that the problem of thermal-based eye detection using still frontal face images is being investigated.
展开▼
