1) We supplied two persons with all iris images extracted from all MBGC video sequences.
Figure 16 is a collection of the six problematic best images using human vision (HBIS) from six (out of the 586) video sequences in the MBGC distant-video dataset.
Using the MBGC dataset, we compared ABIS and HBIS approaches by evaluating the resulting VASIR verification rate (VR).
This thus suggests that VASIR is not only near-unique in that its algorithm automatically extracts the best image from a video sequence, but also automated best image selection (ABIS) of VASIR is almost as good as its (manual) human ground truth counterpart for the MBGC dataset.
More than 45 % of the iris images that were extracted from the MBGC NIR face-visible video dataset are corrupted with noise, are out-of-focus, and have poor contrast.
Figure 28 shows a difference between the Hollingsworth's approach and VASIR's approach--using a Scatter plot of 4,800 points of raw wavelet coefficients from a normalized iris image (240 x 20 pixels)--plotted using one of frames from a video file in the MBGC dataset.
2) All Frames subset from MBGC NIR face-visible video, and
The MBGC NIR face-visible video samples were captured with a video camera by the Sarnoff IOM system.
This MBGC NIR face-visible video dataset consists of video sequences capturing a subject's face while s/he was walking through a portal at normal walking speed from 10 feet away.
2 All Frames Dataset from MBGC NIR Face-Visible Videos