Skip to main content

Iris and Face Recognition

Todaythe science of biometrics is merging with the field of information technology to create new methods for collecting data on a variety of other human physical and behavioral traits as well.  Innovative devices and technologies capable of analyzing such traits as retinas, irises, voice patterns, facial features, and hand measurements are already being used for biometric authentication across a wide array iris recognitionof areas including corporate and public security systems, military surveillance, counter-terrorism initiatives, and point of sale applications. 

Leading the way in the development of new biometric technology is Dr. Marios Savvides, Director of the CyLab Biometrics Lab.  With funding from CyLab, Savvides and his team of researchers are breaking new ground in the enhancement of existing biometric authentication technologies as well as the creation of revolutionary new ones.  Savvides’ work is specifically geared toward improving the use of iris and face recognition in biometric authentication systems. 

Savvides’ work in iris recognition focuses on two problems inherent to its use as a means of biometric authentication:  image quality and device efficiency.  Images collected with iris recognition devices are frequently of low quality, containing random specular reflections in and around the pupil and iris that impact the performance of iris segmentation algorithms.  To improve image quality, Savvides developed a more robust iris segmentation algorithm capable of segmenting an image of the iris in an average time of 3.37 seconds, versus an average time of 15.99 seconds using the old algorithm.  Reduced segmentation time yields fewer instances of specular reflection in iris images and lowers the iris segmentation error rate from 9.65% to 3.66%. 

Savvides has also focused on enhancing device efficiency.  Most iris recognition devices can capture only one image of an iris at a time.  After each image capture, the device user must manually enter several pieces of identifying information, including whether the image is of a left eye or a right eye.  The single capture ability of iris recognition devices slows the data collection process and increases the likelihood that iris images will be misidentified and mislabeled.  In response to these problems, Savvides developed an iris recognition algorithm capable of detecting left eyes from right eyes, allowing devices to capture images of both irises simultaneously.  Tests of the new algorithm on four different iris databases have shown it to be highly efficient, adding no additional computation time to current iris recognition systems and with a 99% identification accuracy rate.

Face Recognition Technology

Savvides has also made great progress toward improving face recognition technology.  Two of the biggest challenges with face recognition are image capture and pose correction.  face recognitionClear facial images are extremely hard to capture when subjects are moving fast in a crowd.  Despite the fact that surveillance video may be able to capture hundreds of frames of a subject’s face, there may only be one or two frames that are actually useable.  Savvides has developed new algorithms for use in face recognition technology which enable cameras to capture 100 image frames per second, thereby increasing the likelihood of pinpointing useable face images.  Another challenge with face recognition technology lies in the processing of non-frontal facial images.  In order for an image to be properly enrolled in a face recognition database, the target subject must be frontally posed, which can be difficult if a subject is uncooperative or if a non-frontal image is acquired from a print source or surveillance video.  To address this problem, Savvides developed a process to correct a non-frontal image into a frontal neutral pose.  The process takes a two-dimensional, non-frontal image, identifies facial feature landmarks, and reconstructs the image into a three-dimensional model using depth mapping techniques.  The extrapolated 3D model is then rotated and pose corrected into a frontal image that is normalized into a 2D image suitable for use in face recognition systems.

Iris and face recognition devices are already being used by the U.S. military to identify and track suspected terrorists and other enemies.  Although such devices have proven extremely helpful to the military, the underlying technology is not without its drawbacks.  Iris scanners, for example, only work when targets are stationary and within very close range, making it impossible to capture iris images from moving targets at a distance.  With support from CyLab and the Department of Defense, Savvides is currently developing enhanced iris and face recognition tools for military and counter-terrorism use, such as iris scanning technology capable of detecting and identifying unique iris markers in moving targets from up to 43 feet away and security camera software able to capture facial images from a distance of up to 30 yards.

As security concerns become more prevalent in every aspect of life in the 21st century, so too will the need for biometric authentication devices that are more effective and efficient.  Evidenced by the pioneering work of Dr. Marios Savvides, the CyLab Biometrics Lab is committed to advancing the state of the art in the vital and growing fields of biometric authentication technology.