IITBBS Cross-Spectral Periocular Dataset

Motivated by the scarcity of cross-spectral periocular datasets containing images in unconstrained environments, we create an in-house dataset, namely, the IITBBS cross-spectral periocular dataset, with VIS and NIR periocular images from 100 subjects aged 20-55 years. Unlike existing datasets, the new cross-spectral dataset contains frontal periocular images as well as those with eye and head movements from left to right and top to bottom directions. The multiple directional movements reflect the unconstrained nature of real-world scenarios where the subjectscould give random pose variations. In recent times, facial masks have become a necessity to avoid hazardous viral infections (e.g. recent COVID-19 outbreak). Similarly, eyeglasses are one of the most commonly used accessories in our day-to-day life. The eyeglasses induce reflections in the periocular region which affects performance of the periocular recognition systems. The use of both mask and eyeglass creates moisture which degrades the performance further. Existing datasets devote a small percentage of their strength towards the images equipped with eyeglasses. Also, the existing cross-spectral periocular datasets do not contain images with masks. On the contrary, we have captured periocular images by incorporating both eyeglass and mask accessories where images from 46 subjects are equipped with eyeglasses and 34 of them are equipped with both mask and eyeglasses. Consent of the individual subjects and institutional permission were obtained to create this new publicly available cross-spectral periocular dataset. A HIK Vision IR cube day and night network camera was employed to create this dataset. This camera is an embedded digital surveillance product with 4 MP resolution, enabling it to capture cross-spectral images with dimensions of 2688\u00d71520 pixels.The device automatically switches the mode of operation between VIS and NIR by detecting the presence of light.