Skip to main content

Data Sets

Multispectral Dataset

Due to the DoD sponsored data collection we are not allowed to distribute the Multispectral Dataset to foreign nationals or researchers outside USA.

ONR Collections – SWIR Biometrics Dataset

Goal: Develop algorithms for cross-spectral face matching at night and obstructed by tinted materials.

WVU Multispectral Face Database

Three different types of cameras have been used to collect this database, i.e. a Canon EOS 5D Mark II, a DuncanTech MS3100 and a XenICS Xeva InGaAs. These cameras were used to acquire the WVU Multispectral database that is composed of three datasets, i.e. visible (CANON), multispectral (MS3100 - captures four different wavelength bands, i.e. IR and Red, Green and Blue), and SWIR (XenICS). The purpose of acquiring such a database was to capture face images under different spectra in order to benefit from the information variability provided by different modalities, and study the recognition performance of different techniques.

The database was collected over two sessions spanning one month. In total, 30 subjects participated, 22 male and eight female. Each session of the Visible dataset has 15 frontal face images for each subject, resulting in 450 images in total. In each session of the Multispectral dataset , we acquired face images at three different poses, i.e. full frontal, and left/right at +/−67.5 degrees. For each pose and with a single shot we captured four images that are associated with the Green/Blue, Red, IR, and RGB/Color-IR (combined) component respectively (360 images in total). For the SWIR Dataset, we acquired face images at three different poses, i.e. full frontal, and left/right at +/−67.5 degrees. For each pose, we obtained images with and without employing a band-pass filter, viz. a set of hardware/lenses that were placed in front of the camera before each shot. The wavelength range covered starts from 900 nm, and goes up to 1700 nm, using filters that are 100 nm wide and centered at 950, 1050, . . . , 1650 nm. Thus, nine samples per pose (no filter, and 950–1650 nm/100 nm) have been acquired, resulting in 810 images in total (27 samples/session×30 subjects).

Multispectral Imaging (Iris) Database

This collection includes the images of human iris. A multispectral image consists of multiple wavelengths or wavelength bands of the electro­magnetic spectrum captured over the same object, independent of the other wavelengths. It is typically represented as a matrix of pixels for each 2-Dimensional channel (eg. Red, Green, Blue, etc.) that are concatenated or stacked on top of each other in the 3rd dimension. These mul­tispectral images are typically a narrow band over certain wavelengths ranging around 50nm in bandpass.

The arrangement described above was used to acquire 70 eye images from the left and right eyes of 35 subjects (5 samples per subject). The subject pool used in this preliminary analysis consisted of individuals having different eye colors. The database includes irides ranging in color from greyish blue, greenish hazel, to dark brown. The ’/’ indicated in the color column of the table suggests that the iris has a two-toned appearance.

The data set also contains distinguishing features such as freckles, moles, and various colored spots and streaks. Only subjects 9, 10, 13, 17, and 35 of the set have irides with little or no melanin pigmentation and 15 out of the 35 subjects have some form of multicolored iris. Thus, the data set is very diverse with respect to the color contained in the irides. The color of each iris was determined by visual inspection since it is difficult to automatically elicit the eye color from the images given the rapid variations in texture chromaticity within the high-resolution image.