This is Dev Technology’s first post in a series of blog posts on emerging biometric technologies. Use this link to see all posts in the series.
The ability to identify a person or distinguish between two different people based solely on the features of face is a skill people use their entire lives. While this is something most people take for granted, reproducing this ability in machines by using a system of cameras, algorithms, and computer power to accurately accomplish facial matching in real-time has proved to be surprisingly difficult for the biometric community.
Initial research focused on the use of cameras utilizing the visible light spectrum since these cameras were the cheapest and have the sharpest resolution for greater picture detail. However, recent technological advantages in infrared (IR) cameras–the next longest wavelength of light past visible light–have lowered equipment cost, increased picture resolution, and pushed research in facial matching in that direction. Infrared cameras have several advantages to visible cameras when it comes to facial recognition, but overall, the technology still faces various challenges.
Advantages of Infrared Facial Matching
Ability to perform facial recognition in the dark: IR cameras and Long Wave Infra-Red (LWIR) imagery is completely independent of the need for target illumination since IR sensors operate by measuring heat energy emitted and not the light reflected off of objects. With no need for lighting, IR cameras can capture detailed images at greater range than a visible camera.
Not affected by varying environmental conditions: IR energy can be viewed in any lighting conditions and is less subject to scattering and absorption by smoke or dust than visible light. As a result, infrared facial recognition produces results in environments and conditions where visual facial recognition would fail.
Less sensitive to head orientation and facial expressions: The position, angle, rotation, and tilt of the face in the captured image can have a large negative effect on facial matching algorithms. Recent research in the field has found that infrared based facial recognition is more invariant than visible light based matching under various conditions, specifically varying head orientation and facial expressions. Facial expressions and head orientation movements cause direct changes in the facial structures of the image, as well as changes to the contours of shadows from the illumination source in visible cameras. In an infrared image, this shadow effect is greatly reduced as no illumination source is needed.
Ability to detect disguises: Disguises such as wigs, fake mustaches, and sunglasses are inexpensive methods to deceive facial matching algorithms using traditional visible cameras. When viewed in the infrared, disguises are easily discerned. Matching algorithms can be modified to detect tell-tale disguise signatures and alert system operators of suspicious behavior if in a controlled environment, or ignore the area of the face where the disguise was found and concentrate on matching other facial features found.
In summary, infrared facial matching is nearly invariant to illumination and facial expression changes between images captured, works in even total darkness, and is useful for detecting disguises. As a result, detection, location, and segmentation of one individual face from multiple faces is easier when using thermal images, and it is also biometrically more accurate then facial matching using visual cameras.
Challenges of Infrared Facial Matching
Despite the benefits of infrared facial matching when compared to facial matching with visual cameras, infrared facial is not without its challenges:
- Accuracy is decreased as the technology is not based on a strong biometric trait such as fingerprints or irises which can even tell identical twins apart.
- Images change when subject inhales and exhales through the nose as the temperature of the air passing through the nasal passage changes the thermal signature of the image. Possible solutions being explored include combining multiple image captures to reduce this effect.
- Metabolism effects from such things as eating, exercise, alcohol and caffeine consumption can cause changes between two captured images.
- Glasses distort image and obscure the details around the eyes.
- While advances have been made in resolution and costs have significantly lowered since the 1990’s, LWIR camera systems are still lower resolution and more expensive then visual cameras.
Research in the field is currently focusing upon these challenges. The advantages of an automated facial matching system for passive surveillance and identification far outweigh the cost and difficulty of the continued research in this field.
About the Author: Mike Wagner is an IEEE Certified Biometric Professional and has performed biometrics development on projects such as US-VISIT and the system used by the Border Patrol to collect biometric data in the field. In addition to his focus in biometrics, Mike Wagner also specializes in enterprise architecture, Java software development, large system troubleshooting and problem solving.