We conducted more than 1.3 million comparisons of iris patterns encoded from images collected at two Nigerian universities, which constitute the newly available African Human Iris (AFHIRIS) database. The purpose was to discover whether ethnic differences in iris structure and appearance such as the textural feature size, as contrasted with an all-Chinese image database or an American database in which only 1.53% were of African-American heritage, made a material difference for iris discrimination. We measured a reduction in entropy for the AFHIRIS database due to the coarser iris features created by the thick anterior layer of melanocytes, and we found stochastic parameters that accurately model the relevant empirical distributions. Quantile-Quantile analysis revealed that a very small change in operational decision thresholds for the African database would compensate for the reduced entropy and generate the same performance in terms of resistance to False Matches. We conclude that despite demographic difference, individuality can be robustly discerned by comparison of iris patterns in this West African population.

This paper discusses the role of symmetry-breaking in biometric recognition. Using publicly available databases, we investigate three kinds of broken symmetries in iris patterns: binocular, monocular, and monozygotic. We report a small but statistically significant difference in similarities between the ipsilateral and the contralateral eyes of twins, and also between genetically identical and nonidentical eyes. Another new finding is a doubling in the variance of Hamming distance scores under a simple monocular mirror transformation, which is consistent with an assessment of entropy.

The discriminating powers of biometric patterns derive from their entropy, just as the hardness of cryptographic keys derive from their entropy. The larger the number of independent bits, or the more independent they are, the less chance of collision. The authors measured the mutual information entailed by radial correlations within each of 632,500 different iris patterns from persons of 152 nationalities. For each iris, they measured how well the sequence of bits in any ring of the IrisCode predicts the sequence of bits in the other rings. Information density is quite non-uniformly distributed across iris patterns radially. Their measurements of mutual information address how much radial resolution is productive to use when encoding an iris, and they show that a non-uniform allocation of encoding resolution radially leads to significant performance improvements by reducing redundancy.

Linear combinations of metrics for assessing biometric sample quality are weak, because they lack veto power. For example, a good score for a sharp focus of an ocular image would `compensate' in an additive combination for the fact that the eyelids are fully closed; or fully open eyelids would compensate for the image being many diopters out-of-focus. Normalised multiplicative quality factors are better because they are punitive, and thereby confer veto powers. This chapter explains the basis for the product of power functions which underlie the ISO/IEC 29794-6 Iris Image Sample Quality Standard, in particular how the exponents of the power functions allow importance tailoring of each element.

The authors generated 316,250 entire distributions of IrisCode impostor scores, each distribution obtained by comparing one iris against hundreds of thousands of others in a database including persons spanning 152 nationalities. Altogether 100 billion iris comparisons were performed in this study. The purpose was to evaluate whether, in the tradition of Doddington's Zoo, some individuals are inherently more prone than most to generate iris false matches, while others are inherently less prone. With the standard score normalisation disabled, a detailed inter-quantile analysis showed that meaningful deviations from a universal impostors distribution occur only for individual distributions that are highly extreme in both their mean and their standard deviation, and which appear to make up <1% of the population. In general, when different persons are compared, the IrisCode produces relatively constant dissimilarity distances having an invariant narrow distribution, thanks to the large entropy which lies at the heart of this biometric modality. The authors discuss the implications of these findings and their caveats for various search strategies, including 1-to-first' and 1-to-many' iris matching.

Highlights? Different images of the same object rarely lead to identical encodings. ? Variations in encodings do not imply variations in the object being imaged. ? A recent paper confounded these and drew strong but unsupported conclusions. ? Algorithmic failure does not justify the inference that the iris has changed.

We investigate three schemes for severe compression of iris images in order to assess what their impact would be on recognition performance of the algorithms deployed today for identifying people by this biometric feature. Currently, standard iris images are 600 times larger than the IrisCode templates computed from them for database storage and search; but it is administratively desired that iris data should be stored, transmitted, and embedded in media in the form of images rather than as templates computed with proprietary algorithms. To reconcile that goal with its implications for bandwidth and storage, we present schemes that combine region-of-interest isolation with JPEG and JPEG2000 compression at severe levels, and we test them using a publicly available database of iris images. We show that it is possible to compress iris images to as little as 2000 bytes with minimal impact on recognition performance. Only some 2% to 3% of the bits in the IrisCode templates are changed by such severe image compression, and we calculate the entropy per code bit introduced by each compression scheme. Error tradeoff curve metrics document very good recognition performance despite this reduction in data size by a net factor of 150, approaching a convergence of image data size and template size.

We investigate three schemes for severe compression of iris images in order to assess what their impact would be on recognition performance of the algorithms deployed today for identifying people by this biometric feature. Currently, standard iris images are 600 times larger than the IrisCode templates computed from them for database storage and search; but it is administratively desired that iris data should be stored, trans- mitted, and embedded in media in the form of images rather than as templates computed with proprietary algorithms. To reconcile that goal with its implications for bandwidth and storage, we present schemes that combine region-of-interest isolation with JPEG and JPEG2000 compression at severe levels, and we test them using a publicly available database of iris images. We show that it is possible to compress iris images to as little as 2000 bytes with minimal impact on recognition performance. Only some 2% to 3% of the bits in the IrisCode templates are changed by such severe image compression, and we calculate the entropy per code bit introduced by each compression scheme. Error tradeoff curve metrics document very good recognition performance despite this reduction in data size by a net factor of 150, approaching a convergence of image data size and template size.

We investigated the randomness and uniqueness of human iris patterns by mathematically comparing 2.3 million different pairs of eye images. The phase structure of each iris pattern was extracted by demodulation with quadrature wavelets spanning several scales of analysis. The resulting distribution of phase sequence variation among different eyes was precisely binomial, revealing 244 independent degrees of freedom. This amount of statistical variability corresponds to an entropy (information density) of about 3.2 bits mm(-2) over the iris. It implies that the probability of two different irides agreeing by chance in more than 70% of their phase sequence is about one in 7 billion. We also compared images of genetically identical irides, from the left and right eyes of 324 persons, and from monozygotic twins. Their relative phase sequence variation generated the same statistical distribution as did unrelated eyes. This indicates that apart from overall form and colour, iris patterns are determined epigenetically by random events in the morphogenesis of this tissue. The resulting diversity, and the combinatorial complexity created by so many dimensions of random variation, mean that the failure of a simple test of statistical independence performed on iris patterns can serve as a reliable rapid basis for automatic personal identification.

We argue that some aspects of human spatial vision, particularly for textured patterns and scenes, can be described in terms of demodulation and predictive coding. Such nonlinear processes encode a pattern into local phasors that represent it completely as a modulation, in phase and amplitude, of a prediction associated with the image structure in some region by its predominant undulation(s). The demodulation representation of a pattern is an anisotropic, second-order form of predictive coding, and it offers a particularly efficient way to analyze and encode textures, as it identifies and exploits their underlying redundancies. In addition, self-consistent domains of redundancy in image structure provide a basis for image segmentation. We first provide an algorithm for computing the three elements of a complete demodulation transform of any image, and we illustrate such decompositions for both natural and synthetic images. We then present psychophysical evidence from spatial masking experiments, as well as illustrations of perceptual organization, that suggest a possible role for such underlying representations in human vision. In psychophysical experiments employing masks with more than two oriented Fourier components, we find that peaks of threshold elevation occur at locations in the Fourier plane remote from the orientations and frequencies of the actual mask components. Rather, as would occur from demodulation, these peaks in the frequency plane are related to the vector difference frequencies between the actual masking components and their spectral centers of mass. We offer a neural interpretation of demodulation coding, and finally we demonstrate a practical application of this process in a system for automatic visual recognition of personal identity by demodulation of a facial feature.