Modalities, Hardware and Software
Much is made about the breadth of biometric modalities, and indeed some of the research into new, more exotic biometrics (ear, gait, odor, etc.) is compelling. But the “big three” modalities that are field-proven and currently in use are fingerprint, face, and iris. Each is currently used extensively in mobile applications. In a mobile context, voice is also important, given the inherent audio capture and playback capabilities of a smart phone. There is no perfect biometric; each has advantages and disadvantages, such as having more reliable matching performance, or being easier to capture. Some mobile approaches are multi-modal, incorporating more than one modality and utilizing “fused” match results.
All biometric solutions include hardware and software components, and many include both client- and server-based components. Mobile biometric hardware includes capture peripherals, power source, network interface, and computing platforms. Mobile biometric software includes a user interface, peripheral interface, biographic data capture and validation, biometric image capture and processing workflow, and biometric template extraction and matching. Mobile capture solutions vary in the degree of required miniaturization and portability, and as such sometimes use the same hardware and software components as for stationary solutions.
Fingerprint sensors are currently designed upon either one of two technologies: optical and capacitive. Capacitive sensors can be either full-finger or swipe. It is important for fingerprints to be of sufficient resolution (500 ppi) and contrast, and be free of distortion. An optical sensor uses a prism, light source, and light sensor to capture images of fingerprints. Capacitive sensors are based on a silicon chip that detects electrical currents when the finger ridges make contact. Optical sensors generally provide higher-quality images than capacitive sensors but are also larger and consume more power. Full-image capacitive sensors generate higher-quality images than swipe sensors but are also larger. Swipe sensors do not generate image quality sufficient for one-to-many identification.
Capture of facial images has traditionally been performed using off-the-shelf consumer-grade digital cameras such as a Canon PowerShot. But camera technology has changed dramatically in only the last several years, making facial capture with mobile devices far more viable. We have all seen the vast improvements in image quality of web cams and smart phone cameras, many of which are now capable of eight megapixels or more. This is compelling, considering that the best digital cameras on the market were on the order of four megapixels only five years ago. Digital facial images traditionally require an interocular resolution of about 60 pixels for one-to-one matching and 90 pixels for one-to-many matching. But resolution is not the only factor affecting facial matcher performance; perhaps even more important are the distortion, brightness, contrast, sharpness, and background clutter of the image. Improvements in these areas have not been as dramatic with smart phone cameras as resolution, but improvements are nevertheless compelling. But the challenge here is less with the camera performance and more with the fact that with mobile solutions, the capture conditions are highly variable as compared to stationary environments; lighting, background, and distance to the subject can change from photo to photo, and have a substantial impact on matching performance. In mobile facial image matching systems, it is helpful to keep photo capture conditions as consistent as possible.
Iris is probably the fastest growing biometric modality, and has also benefitted from the dramatic changes in the camera and sensor arena. But iris differs from face in that it requires an infrared image of the iris. The degree to which a pure infrared image can be captured (with minimal “pollution” from visible light), the better the matching performance. This is why off-the-shelf cameras aren’t used for iris image capture, and a special camera is required; a system must illuminate the iris with infrared light and then filter out other wavelengths.
Voice biometrics are a particularly viable means of one-to-one verification using a smart phone, given its obvious audio capabilities. But in this application mobile voice biometrics suffer from the same challenges as other biometrics in that the environment is unpredictable; background noise can interfere with the matching process just as the background of a facial image can. Voice samples (such as those collected from a surveillance device) can be used as a “latent” like fingerprints, and the background of voice signals can also have forensic value for law enforcement and military applications.
A different take on mobile biometrics is “biometrics in motion,” where biometrics are taken of an individual in motion using stationary equipment. There are several systems available for this application and are desirable for their speed and low degree of interference with a person’s activity. Facial and iris biometric systems are available, and fingerprint systems are currently in advanced research stage.
Hardware challenges: durability, ergonomics, power consumption
Capture hardware must be designed to accommodate a wide variety of environmental and ergonomic factors particular to mobile devices used outside an office environment. They must operate in direct sunlight, temperature and humidity extremes, and by operators with gloved fingers. They must be sufficiently durable to withstand moisture, dust, dirt, and impact. Power consumption is a significant factor and is very use-case specific, but typically a device must operate reliably by battery power for at least one work-day. A consideration is whether batteries can be replaced on-the-fly, or whether the entire device must be offline during a battery charge. Biometric capture peripherals often require lighting, and one-to-many search software can consume processing. They must be ergonomically designed such that an operator can quickly and reliably collect high-quality biometric samples from inexperienced or even uncooperative individuals.
Software for mobile capture is functionally similar to stationary biometric solutions, but with meaningful differences. Mobile devices typically run operating systems designed for smaller devices, such as Windows CE, Windows Mobile, Apple iOS, Blackberry, and Android, so biometric software libraries need to be ported to these operating systems. Processors are less powerful, and there is less memory and RAM, so processing-intensive operations such as compression and template extraction must be optimized appropriately. Video screens are smaller and utilize touch screens and gestures, so the user interface must be designed to accommodate these. Special software is applied to help make up for shortcomings of the hardware, such as by providing more advanced image processing and quality control. Some more powerful mobile devices such as those used in the military run Windows XP but still need software that accommodates a small touch screen and less processing power.
Device independence is a particularly important consideration in implementing a mobile system. Ideally, an owner of a system can support different kinds of mobile devices in their system so that they are not reliant on a single vendor for devices. This can be achieved by requiring that an open, standards-compliant interface be implemented between the device and wherever the trusted sample is stored (i.e. a smart card or central server). Another approach is to operate a hardware-agnostic software application on the mobile device. The advantage here is that different hardware devices can be used in the system and can be procured separately, but the software, user interface, workflow, etc. are the same.