Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
ISO/IEC 29145-2:2022 — Presentation Attack Detection — Part 8: Iris
Technical deep dive into iris liveness detection and presentation attack resistance
Introduction to Iris Presentation Attack Detection
Iris recognition is widely regarded as one of the most accurate biometric modalities, with false match rates as low as 1 in 10 million under controlled conditions. However, the iris is not immune to presentation attacks. Attackers may present printed images of an iris, display iris images on mobile devices, wear textured contact lenses imprinted with a target’s iris pattern, or even use prosthetic eyes crafted to reproduce a specific iris texture. ISO/IEC 29145-2:2022 provides the standardized framework for detecting such attacks against iris recognition systems.
Textured contact lens attacks represent the most practical and concerning threat to iris recognition systems. Commercially available cosmetic contact lenses can be custom-printed with any iris pattern at resolutions approaching 200 DPI — sufficient to fool many commercial iris matchers. The standard addresses this attack vector through both image quality metrics and dedicated lens detection algorithms.
The standard categorizes iris presentation attacks into several classes. Printed iris attacks use high-resolution printers to reproduce an iris image on paper or similar media. Display-based attacks present iris images on electronic screens including smartphones, tablets, and e-ink displays. Contact lens attacks use textured or patterned cosmetic lenses to overlay a target iris pattern onto an attacker’s genuine iris. Prosthetic eye attacks involve artificial eyes that incorporate a printed or manufactured iris texture. Post-mortem iris attacks use iris tissue from deceased individuals, exploiting the fact that iris texture persists after death for a limited period.
Iris Liveness Detection Techniques
Pupil Dynamics and Natural Response Analysis
The living iris exhibits characteristic pupillary responses to light stimulation that are difficult to replicate artificially. The standard describes protocols for measuring pupil light reflex (PLR) latency, constriction amplitude, and recovery kinetics under controlled illumination changes. Natural pupil dynamics include hippus — the spontaneous, small-amplitude oscillations in pupil diameter that occur even under constant illumination, driven by autonomic nervous system activity. An artificial iris or printed image cannot reproduce these dynamic physiological responses.
Iris Tissue Textural Analysis
High-resolution iris images reveal characteristic texture features at multiple spatial scales that differ between live iris tissue and printed reproductions. Live iris tissue shows a three-dimensional crypt, furrow, and collarette structure with depth variations that printed media cannot replicate. Frequency-domain analysis using 2D Fourier transforms and wavelet decompositions can identify the periodic dot-matrix patterns characteristic of inkjet and laser printers. The standard specifies minimum image resolution requirements (typically 200+ pixels across the iris diameter) for reliable textural analysis and provides reference datasets for algorithm training and evaluation.
| Detection Technique | Attack Types Detected | Key Performance Factors | Implementation Considerations |
|---|---|---|---|
| Pupil light reflex | Printed, display, prosthetic | Latency, amplitude, recovery slope | Requires NIR illuminator modulation |
| Hippus analysis | Printed, display, prosthetic | Oscillation frequency 0.05–0.3 Hz | Requires 5–10 second observation window |
| Print detection (frequency domain) | Printed, some display | Dot matrix periodicity detection | Computationally lightweight, single frame |
| Contact lens detection | Textured lens | Edge artifacts, surface reflections | Resolution-dependent, requires iris ROI ≥ 100 px |
| Multi-spectral iris imaging | Printed, lens, prosthetic | Spectral reflectance differences | Requires dual-band (NIR + visible) camera |
| 3D iris shape analysis | Printed, display | Iris curvature and anterior chamber depth | Requires stereo or structured light imaging |
Pupil light reflex testing offers one of the most reliable yet simple-to-implement iris PAD techniques. By modulating the NIR illumination intensity and measuring the temporal pupil response with the existing iris camera, no additional hardware is required. The key engineering challenge lies in robust pupil tracking under varying illumination and off-axis gaze conditions.
Multi-Spectral and Depth-Based Iris Analysis
Live iris tissue exhibits distinct spectral reflectance properties across visible and near-infrared wavelengths. The melanin concentration in the iris stroma creates characteristic absorption patterns that differ from printed inks or contact lens materials. Multi-spectral imaging using both NIR (700–900 nm) and visible wavelength bands can discriminate between genuine iris tissue and artificial reproductions. Additionally, the three-dimensional structure of the anterior eye — including the curvature of the iris plane relative to the cornea and lens — can be measured using optical coherence tomography or Scheimpflug imaging to detect planar printed materials or artificial lenses.
Engineering Design Insights for Implementation
Iris PAD implementation requires careful balancing of detection accuracy, capture time, user convenience, and hardware cost. The standard provides a comprehensive evaluation framework that includes multiple attack species within each attack class, multiple presentation instruments (e.g., different printer models, contact lens brands), and varying capture distances and angles. APCER and BPCER must be reported separately for each attack class to identify modality-specific vulnerabilities.
Contact lens detection remains one of the most challenging PAD problems for iris recognition. Modern cosmetic lenses are designed to be comfortable and transparent — properties that also make them difficult to detect. The lens edge, which typically extends 1–2 mm beyond the limbus, can be identified through high-resolution imaging, but at standard capture distances (20–40 cm) the edge may fall outside the imaging field or be too poorly resolved for reliable detection. Multi-spectral approaches that exploit the different spectral transmission of lens materials compared to corneal tissue offer a promising alternative.
Environmental illumination presents a particular challenge for iris PAD. Ambient visible and NIR light can affect pupil size, introduce specular reflections that mimic lens artifacts, and alter the apparent spectral characteristics of both genuine and attack presentations. The standard mandates PAD evaluation under at least three illumination conditions: low ambient (< 10 lux), typical indoor (100–500 lux), and bright outdoor (1000+ lux) to ensure robustness across deployment environments.
From a system architecture perspective, the standard recommends that iris PAD be integrated into the image quality assessment pipeline that precedes feature extraction and matching. This placement allows poor-quality or potentially attack images to be rejected without consuming comparison computational resources. The PAD decision should be reported as a confidence score that can be thresholded independently of the matching score, giving system integrators flexibility in balancing security and convenience.
Frequently Asked Questions
Q: Can a high-quality printed iris photo fool an iris recognition system?
A: A printed iris photo may fool a system that relies solely on iris texture matching without liveness detection. However, the absence of pupillary light reflex, the flat planar geometry, and the characteristic dot-matrix print patterns all provide cues that a properly implemented PAD system can detect with high reliability.
A: A printed iris photo may fool a system that relies solely on iris texture matching without liveness detection. However, the absence of pupillary light reflex, the flat planar geometry, and the characteristic dot-matrix print patterns all provide cues that a properly implemented PAD system can detect with high reliability.
Q: How effective are textured contact lens detection algorithms?
A: Modern contact lens detection algorithms achieve APCER below 5% at BPCER of 1% under controlled conditions, but performance degrades at longer capture distances and in the presence of motion blur. The standard calls for continued research investment in this area as lens manufacturing quality continues to improve.
A: Modern contact lens detection algorithms achieve APCER below 5% at BPCER of 1% under controlled conditions, but performance degrades at longer capture distances and in the presence of motion blur. The standard calls for continued research investment in this area as lens manufacturing quality continues to improve.
Q: Does ISO/IEC 29145-2 cover infrared iris imaging specifically?
A: Yes, the standard specifically addresses NIR iris imaging (700–900 nm), which is the dominant wavelength band for iris recognition. Test protocols for multi-spectral PAD that combine NIR and visible channels are also defined.
A: Yes, the standard specifically addresses NIR iris imaging (700–900 nm), which is the dominant wavelength band for iris recognition. Test protocols for multi-spectral PAD that combine NIR and visible channels are also defined.
Q: What is the minimum iris image resolution required for PAD?
A: The standard recommends a minimum iris diameter of 200 pixels in the captured image for reliable PAD feature extraction, consistent with the international standard for iris image quality (ISO/IEC 29794-6). Lower resolutions may still support PAD but with reduced accuracy.
A: The standard recommends a minimum iris diameter of 200 pixels in the captured image for reliable PAD feature extraction, consistent with the international standard for iris image quality (ISO/IEC 29794-6). Lower resolutions may still support PAD but with reduced accuracy.
