Biometrics Enhancements
The latest technologies are advancing
biometric solutions for in-the-field identification.
The U.S. Army Materiel Command Information Technology, E-Commerce and Commercial Contracting Center West (ITEC4-W) has a requirement for 3-D facial recognition and multispectral fingerprinting technology for the U.S. Army’s Language and Technology Center at Fort Huachuca, Ariz. The mission of the Language and Technology Office (LTO) is to “demonstrate technological, concepts and architecture paths to reduce language barriers reduced by operational and front line intelligence personnel, as well as the intelligence community and to provide research and development arm of biometric application in support of the global war on terrorism.”
The Biometrics Automated Toolset (BAT) is government off-the-shelf software initially developed in the Battle Command Battle Lab (Huachuca). It integrates commercial off-the-shelf hardware and software into a biometric enrollment and biographical data repository system.
The system currently incorporates three biometric capture devices: facial recognition, iris recognition and fingerprint recognition. The system is used to enroll detainees, human intelligence sources, host national workers, and other persons of interest. There are approximately 2,300 BAT client systems deployed worldwide (2,000 in operations Iraqi Freedom and Enduring Freedom), networked in a server client configuration in which data is shared among BAT systems as well as national level databases. As part of ongoing nation building efforts dictated by these two operations, DoD has an acute need to achieve identity dominance, that is, the capability to conclusively identify detainees, intelligence sources, host nation worker candidates and other persons of interest.
FACIAL RECOGNITION
The first requirement of this acquisition is the development of software that shows great promise is converting twodimensional (2-D) images to three-dimensional (3-D), which greatly enhances recognition accuracy. The development piece of this effort is to enhance the functionality of taking the 2-D image captured at any angle and converting it to 3-D as well as integrating the software into BAT. The identity dominance would be enhanced because one could capture faces at any angle and be able to recognize them as well as capturing faces in the crowd.
The requirement for enhanced facial recognition capability was specified as a joint urgent operational needs statement by the Central Command in September 2006. Specifically cited was that currently deployed technology of face recognition using eigenfaces and principal component analysis (PCA) for the algorithms discriminating between face classes in 2-D space is inadequate. Current technology measures various points on the face and is sufficiently accurate only when the pose is straight on. Any rotation of the head left and right (the y axis) or any up/down variation (the x axis) or tilt (the z axis) to the straight on pose degrades the accuracy for two-dimensional capture. In short, currently deployed 2-D eigenface/PCA capture is sufficiently accurate only when the x=0, y=0 and z=0, that is, when the target face is captured straight on frontal.
Basically, a face that is straight on frontal (0, 0, 0) pose is an ideal controlled image of the face. Controlled facial imagery also includes controlling the lighting so that the face is well illuminated with no shadows, controlled backgrounds that do not contain any other images or imagery noise, and controlled size ratio of the head within the photo or video frame. Face recognition algorithms available today require highly controlled images in order to achieve acceptable identification performance.
The proposed solution is to incorporate 3-D manipulation to the 2-D capture. Three-D manipulation would normalize captured faces where the pose is somewhat deviated from straight on, either rotated left or right or tilted up or down. The normalization process manipulates the actual captured image so as to effect x=0, y=0, z=0. Once normalized via the 3-D software, facial recognition software can be effectively applied to identify the face. This 3-D normalization process facilitates covert applications in which faces are captured unknowingly, and/or at less than ideal angles (uncontrolled imagery). Three-D normalization also facilitates success rates of facial capture and recognition normally degraded by environmental conditions such as high ambient light as well as physical features partially obstructing the face, such as a branch. The matching algorithms which benefit from the normalization process must also take into account other 3-D information on the face, such as curves and volumes, to provide a more accurate and robust identification result.
MULTISPECTRAL FINGERPRINTING
The second requirement of this acquisition is the development of multispectral fingerprint technology. Quality digital fingerprint capture is adversely affected by environmental factors such as moisture and high ambient light, conditions that exist in the field. Adverse environmental conditions affect the current technology because the finger has to actually make contact with the glass platen on the device. Multispectral fingerprint technology is touchless; moisture and background light do not matter. This technology promises higher quality fingerprint image capture under a wider range of environmental conditions.
Current fingerprint readers are prone to poor image quality under a variety of common conditions. Small amounts of contamination or excessively dry or moist skin can hamper or even preclude the capture of an acceptable fingerprint image. Further, a significant percentage of the population has fingerprints that result in consistently poor images because of genetics, poor health, age or even profession. Ridges are sometimes worn, or they don’t have enough collagen to remain firm when pressed on the platen. Conventional optical sensors are also subject to being “flooded” in direct sunlight and typically cannot handle extremes in temperatures.
Unlike conventional readers, multispectral imaging technology captures data from the surface and the subsurface of the fingertip, resulting in superior fingerprint images over a wider range of adverse environmental conditions. Thus, multispectral sensors are not dependent on perfect or unobstructed contact with the sensor, quality of ridges or optimum conditions.
A rugged, portable, compact, easy-tooperate, affordable fingerprint scanner that solves problems with image capture is urgently required in field operations. Force protection, civilian disaster assistance, border security, law enforcement and intelligence collection efforts would be greatly enhanced. The proposed solution is needed by the warfighter and would be employed in the concept of operation of numerous battlefield operating systems, including military police, military intelligence and special operations forces. This solution, when deployed, will be a significant enhancement to efforts in the GWOT. ♦