Pupil detection under lighting and pose variations in the visible and active infrared bands

We propose a novel and efficient methodology for the detection of human pupils using face images acquired under controlled and difficult (large pose and illumination changes) conditions in variable spectra (i.e., visible, multi-spectral, and short wave infrared (SWIR)). The methodology is based on template matching, and is composed of an offline and an online mode. During the offline mode, band-dependent eye templates are generated for each eye from the face images of a pre-selected number of subjects. Using the eye templates that are generated in the offline mode, the online pupil detection mode determines the locations of the human eyes and the pupils. A combination of texture- and template-based matching algorithms is used for this purpose. Our method achieved a significantly high detection rate, yielding an average of 96.38% pupil detection accuracy across all datasets used. Based on a comparative analysis on different databases, we concluded that: (i) a single methodological approach can be used to efficiently detect human eyes and pupils of face images (with strong pose and illumination variations) acquired in the visible and hyper-spectral bands, and (ii) the use of texture-based matching and normalized band-specific templates significantly increases detection accuracy. To the best of our knowledge, this is the first time in the open literature that the problem of multi-band pupil detection on face images in the presence of lighting and pose variations, is being investigated using a unified algorithm.