The implementation of an iris biometric system has three major components: the image acquisition device; the biometric software for template creation, enrollment, and matching; and the database management platform...
Iris recognition is catching up to other popular biometric applications, such as fingerprint and facial recognition, in worldwide usage. It’s a highly accurate technology because human iris patterns don’t change with age and are more challenging to counterfeit. However, the iris’ qualifying image is also more challenging to capture than a face or fingerprint.
Generally speaking, the implementation of an iris biometric system has three major components: the image acquisition device (the iris camera); the biometric software for template creation, enrollment, and matching; and the database management platform commonly used in passport control, airport kiosks, access control, and law enforcement applications (Figure 1). This article focuses on the design of the image acquisition device.
Figure 1 An iris biometric system has these three major building blocks. Source: Videology Imaging Solutions
Let’s look at some of the key challenges in the design of a capture device for iris recognition.
The larger the distance from the camera to the subject’s eyes, the more complex and expensive the capture device will be. Consequently, iris cameras are divided into categories based on their capture range. Here, 10-30 cm is the most common range. The image sensor characteristics and the modulation transfer function (MTF) of the system are critical to further subcategorize the camera as means for enrollment and/or verification. The former requires a higher MTF.
Variety of eye colors
Since image contrast is used to extract the patterns of the iris, the capture light wavelength has to support a broad range of colors. Thus, the spectral response of the sensor and the illumination wavelength must be chosen with considerations for power consumption, IEC-62471 compliant eye safety radiation, and synchronization of the pulsing light with the integration time of the camera.
The device has to tolerate some degree of target motion and that’s where the capture volume designed in the optics of the device plays an important role, particularly the depth of field of the camera. The aperture of the lens, the amount of light radiated during capture time, the light wavelength, and the properties of the lens determine the depth below and above the target distance where the camera stays in focus.
Match-ability and interoperability
A good iris image acquisition metric compares images of the same subject captured with the same camera and computes the percentage of images that successfully match. This concept is known as match-ability. When images from the same subject taken with cameras from different vendors are matched, the concept is known as interoperability.
The percentage of images that match is an indicator of the success of the camera under test to work along other iris cameras in the same system. Meanwhile, the hamming distance between images is the metric used to measure match-ability. The hamming distance has a range from 0 to 1. The closer to 0, the more identical the images are. The camera settings for image enhancement are tuned to set the best conditions under which the captured images achieve high scores of match-ability and interoperability.
Iris recognition standards
There are two ISO standards that set the requirements for iris image quality and exchange of iris image information: ISO/IEC29794-6 and ISO_IEC_19794-6. There is no official biometric standard for iris camera selection equivalent to, for example, the FBI’s Appendix F for fingerprints. However, NIST provides guidance to standardize and develop this technology through a significant number of publications and studies.
A key component of iris recognition is, of course, the biometric software that has the algorithms to extract the unique patterns from the iris image to create an iris template. The iris template is the instrument used for enrollment and matching. Currently, template creation is proprietary, meaning that a template file created with one biometric software can only be encoded and decoded by that software package, which is exclusive to the company that owns the software.
Algorithm development for iris image quality grading, template creation, enrollment, and matching is a whole field in itself. Some companies do both the image acquisition device and the biometric software, while others do one of them. As this technology keeps growing, the opportunity expands for image acquisition devices and access control embedded platforms that can work with multiple third-party biometric software products.
The iris camera requirements for enrollment and verification (matching) is different. The ISO standards mentioned earlier do apply to enrollment cameras; however, verification cameras are not required to follow the standards. Aiming to develop a low-cost enrollment camera that does both—enrollment and verification—offers an alternative to the integrators of iris recognition solutions because they can then architect a solution with iris cameras and a software platform that comes from different vendors.
It would be a budgetary solution for the small integrators who seek access control for a small factory or school. At the same time, it would be a viable procurement strategy for a large integrator, such as the government or the military.
The main task of the acquisition device is to locate the iris of a person within a live video stream and to deliver it to the biometric software for segmentation, which, in turn, creates an iris template for either enrollment or verification. Consequently, in addition to the capture volume specifications, the acquisition device’s on-board processing has to perform a combination of tasks at runtime, which can be carried out by an Arm processor, an FPGA, or an embedded Linux board. That’s how some self-contained systems function in access control applications.
An FPGA-based solution
Here is an example of the iris recognition tasks involved in an FPGA-based solution.
This determines which frames are worth processing while discarding the rest.
There are various techniques to carry out this task. A popular one is to use a circular edge detector to locate the coordinates of the iris within the image.
Image cropping for segmentation
Once the coordinates of the iris are located, the image of the corresponding eye is cropped from the full frame and delivered to the biometric software for image quality assessment and subsequent template creation.
Figure 2 This diagram shows how an FPGA board performs the iris acquisition process. Source: Videology Imaging Solutions
Parallel computations, control of the video pipeline, and power consumption are important considerations for analyzing the image. Some iris cameras use a multi-camera solution, in which one camera looks at the face to locate the eyes within the capture volume, which then triggers another image sensor for the iris capture. In contrast, other cameras use only one sensor for iris capture.
These choices depend on the design tradeoffs and camera category that a designer wants to target along with other design considerations such as distance range, enrollment vs. verification, and software development kit (SDK) integration.
In a recent national biometric rally conducted by the Department of Homeland Security, the whole transaction time—from the moment a person stands in front of the iris camera to template creation of both irises—was targeted to no more than 8 seconds, with a goal of no more than 5 seconds. There are external factors that affect the transaction time, some of which are out of the system’s control, including eye occlusion, gaze rotation, a person’s height, and eye diseases. Therefore, the acquisition device should analyze the images very fast, if possible, so that the biometric software doesn’t get saturated with useless samples.
Board designs for iris recognition
The acquisition device is fundamental to the iris recognition process. Therefore, as the iris recognition industry gains traction in the biometrics field, the opportunity grows for accurate and cheaper acquisition devices that can carry out the search for the coordinates of the iris and apply image enhancement techniques on-board.
At the same time, however, the robustness of the iris biometric software is crucial to grading, segmenting, and matching an iris image. Through the development process, an iris biometrics software package with multiple grading parameters is critical, allowing the acquired images to be analyzed and tuned to achieve the highest possible image quality grade, consistent iris dilation, consistent focus, and high contrast.
There is simply no substitution for speed. The speed at which the camera locates the coordinates of the iris and delivers the iris image to the biometric software for segmentation is no less critical than the quality of the captured image.
W. Luis Camacho is a senior electrical engineer at Videology Imaging Solutions and the architect of the IDentity-1 iris scanner.