Ergonomics, Robustness to Noise, and Imaging Factors

In many respects the iris of the eye is inherently difficult to image at a comfortable ``social" distance (e.g., several feet from a mounted videocamera). It is a small tissue only 11mm in diameter, and hence optical zoom is required, which creates problems of target motion amplification and limited depth of field for focus. More critical even than these limitations of spatial resolution is the limitation of grey-scale resolution, since without appropriate gain control of the video signal, many very darkly pigmented irises tend to be digitized flatly into only the lowest few states of an 8-bit A-to-D converter and thus reveal little structure. A further reason that spatial resolution is less of a challenge than grey-scale resolution is because the upper roll-off frequency of the multi-scale bandpass 2-D Gabor encoders can be equated to a ``blur circle" always larger than three pixels in diameter, which effectively makes any spatial resolution sharper than this irrelevant. Significant parts of the multi-scale iris code are based on analysis of the coarser modulations of this mottled tissue; indeed, some of the 2-D Gabor encoders that are deployed subtend as much as a 70° angle around the pupil. In addition to these issues of resolution, a further challenge arises from the fact that unpredictable amounts of the iris may be occluded by eyelids or corrupted by random silhouettes of the eyelashes.

All of these factors contribute to the observation that different images of the same eye at different times may generate iris codes that disagree in as many as 25% of their bits (the highest observed Hamming distance in Figure 10, for ``Authentics"). This percentage would be the net result, for example, if only half of the bits were deterministic and matched perfectly, while the entire other half were completely random and hence agreed just by chance half the time, yielding an overall agreement of 75% and thus a 0.25 normalized Hamming distance. The robustness of the present recognition method under such high levels of pattern degradation, noise, and inherent imaging limitations, is only possible because of the high statistical complexity associated with the myriad degrees-of-freedom in the iris signal. It is the consequent narrowness of the distribution of Hamming distances for unrelated eyes (the ``Imposters" black distribution shown in Figure 11) that makes any Hamming distance significantly lower than 0.35 virtually impossible to achieve from independent random processes, i.e., unrelated eye images. Thus the hypothesis of independence can be strongly rejected over all but a narrow range of possible Hamming distances.

It is perhaps illuminating that at the ``cross-over" Hamming distance of 0.321, at which point confidence against both types of errors is better than 1 in 10⁵, the level of image degradation or mismatch that is tolerated would be equivalent to obscuring fully two-thirds of the iris (producing just chance 50% agreement among those bits) while finding complete agreement among the remaining one-third of the bits. This extreme example illustrates the robustness against occlusion and noise that can be achieved by converting a pattern recognition problem into a test of statistical independence with a sufficiently large number of degrees-of-freedom.