No change over time is shown in Rankin et al. "Iris recognition failure over time: The effects of texture"

The recently published paper by Rankin et al. [1] ‘‘Iris recognition failure over time: The effects of texture’’ is paradoxical because in fact no changes over time were demonstrated, either in the iris patterns or in the system performance. The authors measured iris recognition performance using ‘‘local and non-local dipole and tripole’’ methods of iris feature extraction proposed by Sun et al. [2] at two points in time: namely after an interval of three months, and after six months, from initial enrolment; but not after zero interval. The performance of their multi-pole algorithm implementation was terrible at both of these time intervals, namely a 20% failure rate. But it did not change over time. The performance of their system was just constantly bad, probably because it ignored small head tilts or eye cyclotorsion, or because it gave unstable segmentation of the iris boundaries, all of which cause shifts or deformations in coordinates and therefore high dissimilarity scores. Nonetheless the authors assert that their study measures ‘‘changes in iris texture patterns over a short time period’’ (p. 145). It appears the authors have just assumed that a non-zero Hamming distance (defined as the fraction of bits that disagree between two iris codes) signifies a change in the iris pattern. But such scores commonly arise simply from algorithm weaknesses such as unstable coordinate alignments. The paper does not indicate that the authors compensated for small variations in head or eye tilt (roll angle) in the matching process. Suppose that a subject’s iris has (say) a dark furrow in the 12-o’clock position. For a dipole feature detector centred exactly over this feature, a tiny tilt of the head or eye clockwise will set its bit one way, while a tiny tilt counter-clockwise will flip the bit to the opposite value. Likewise, small variations in isolating the boundaries of an iris in different images cause shifts or deformations in the coordinate system, again causing bits to flip in its iris code. The authors chose to enforce circular models for these boundary contours, but such models can generate rivalrous alternate solutions. In the absence of careful attention to these issues in an iris recognition algorithm, it is easy to get Hamming distances as large as 0.25 from same-eye (same day) images. Such errors are greater for fine textures than for coarse textures, consistent with the authors’ finding, because any given spatial offset is a larger phase shift for high spatial frequencies than for low. But to ascribe such effects to ‘‘features that change’’ in the iris and to ‘‘variations detected in images captured over time’’ (p. 149) and to ‘‘changes in iris texture’’ (p. 145 and throughout the paper) when mere algorithm implementation failures are responsible and constant over time, is to commit a major, yet obvious, logical fallacy. It is unfortunate that despite these repeated assertions of ‘‘changes in iris texture appearance,’’ no such photographic evidence was provided in the paper. In any case the authors used illumination in the visible band (400–700 nm) which would detect pigmentation changes, unlike illumination in the infrared band, used in all deployed iris recognition systems. Apart from pupil dilation and constriction (which algorithms compensate for with elastic models) and apart from the effects of trauma, botched surgery, and rare disease-related adhesion of the iris to the lens at points (‘‘synechiae’’), there is no published documentation of changes in iris texture, and importantly not in the infrared band. Indeed a recent clinical study [3] conducted in that waveband using a commercially deployed iris camera concluded that ‘‘performance of iris recognition was remarkably resilient to most ophthalmic disease states’’. Some pharmacological treatments for glaucoma involving prostoglandin analogues such as latanoprost are reported to affect iris pigmentation (melanin) when applied topically to the eye [4]. Similarly freckles, pigment blotches and colour changes can develop over time, especially in early childhood. Such possible changes in iris