Face Recognition

This research is being led by Mahesh Babu, Shimon Modi and Eric Kukula

A number of research projects have been undertaken related to testing and evaluating face recognition systems (FRS). Past and present research in this area explores the effects of various extraneous variables on the quality of the acquired biometric data (which are still images or video sequences that include an individuals face) and the way the FRS matcher performs. The extraneous variables may be environmental, contextual or technology related.

Current Research Projects (Spring 2007)

Title: The Effect of Camera Technology on FRS Performance

Researcher: Mahesh Babu

Abstract: Current testing and evaluation standards for the performance of face recognition algorithms do not address the effect of the quality of the sensor or capture device. Standardized testing protocols do not incorporate sensor quality as one of the factors to be analyzed when benchmarking face recognition algorithms. The impact of camera sensor quality on FRS performance has never been examined. In addition, no formal test design and protocol currently exists for examining this impact. The intent of this study is to examine how using different quality sensors affect the capabilities of a face recognition algorithm. The study will also explore the effect of using different quality sensors for enrollment and recognition on the performance of FRS.  

***LOOKING FOR PARTICIPANTS***

Interested in being involved in this study? We are actively looking to recruit participants to be involved with our data collection process. Please visit the Registration page to sign up.

Title: The Effect of Artificially Illuminating Facial Images on FRS Performance

Researchers: Shimon Modi, Mahesh Babu

Abstract: The purpose of this research is to examine the effects of artificially illuminating captured facial images in order to optimize facial recognition performance. In order to study the effects of artificial illumination facial images taken at a low lighting level will be subjected to varying levels of illumination using image processing tools. The difference in image quality and false match and false non match rates will be examined to determine its effect. VeriLook SKD and Identix Face Recognition Engine will be used for matching purposes.

Previous Research

Fall 2006

Title: The Effect of Artificially Modifying Background Color on FRS Performance                                                                   

Researchers: Dr. Elliott, Mahesh Babu

Student Researchers: Mike Muller, Anthony Campbell

Abstract: This study examined whether artificially modifying a facial image using image editing software tools has an impact on its quality or on FRS performance. The images used for this study were originally captured with a uniform 18% gray background. Adobe Photoshop CS2 was utilized to modify the background color in each of the images and create two sets of artificial images: one set with a light blue background and one with a bright green background.  The study showed that there potentially was a statistically significant difference in match scores when the background color was artificially modified. Future work in this area would involve examining the effect on FRS performance of artificially modifying different image attributes, such as illumination levels (see above) and facial characteristics.

Title: The Effect of Facial Image Background Color on FRS Performance

Researcher: Mahesh Babu

Student Researchers: Will DeLozier, Dennis McAndrews

Abstract: Current research in face recognition has not addressed the effect image background color has on the accuracy and precision with which the FRS performs matches. While the current data capture and storage standards for face recognition (INCITS 385) recommends a uniform 18% gray background to optimize FRS performance. On the other hand, current identification schemes that use facial information (such as passports and drivers licenses) either require different color backgrounds or only specify background uniformity and not color. This study intended to explore the rationale behind this disparity and whether it would have an impact on FRS performance. Preliminary results from the study show that background color did not have an impact on image quality and the performance of the FRS system. Future work in this area would involve replicating this study using FRS systems that use different face detection and recognition techniques.

Older Research Projects

Spring 2006

The study examined the perception of individuals to their photograph, and whether their highest ranked self selected photograph had the highest image quality as determined by two face image quality algorithms. The research question was to determine if people, when given a choice, will select the best quality photograph to submit to the passport or consular (or other authorities). The study showed that people do not self-select their best quality images. There was no statistically significant relationship between an individual’s perception and a software based perception of image quality.

2003-2004                                                                   

Researcher: Eric Kukula

Previous studies have examines the relationship between the performance of a face recognition algorithm and illumination levels. The population used consisted primarily of 18-25 year olds. Each subject was enrolled in each illumination condition and was photographed in each of the 3 different illumination conditions over a 4 week period. The results indicated that illumination affected performance of the face recognition algorithm when level I (7-12 lux) and level II (407-412 lux) enrollments were compared against all three levels (level I, II, and III). Furthermore, the results of the level I, II, and III verification attempts using the level III (800-815 lux) enrollment were not statistically significant. As face recognition continued to develop, this protocol was adapted to evaluate a 3D face recognition system. Much like the initial study, this project examined the effects different illumination levels have on the performance of a 3D face recognition algorithm. In addition to illumination levels, illumination directions were also evaluated. Again the population consisted primarily of 18-25 year olds. One main difference between the two studies was in the image acquisition process. This study followed the guidelines in INCITS 385-2004 for capturing images for face recognition systems. Another change from the initial protocol was in enrollment instructions. Each subject was enrolled once in an illumination level of 220-225 lux, which was used for all verification attempts at enrollment/level I (220-225 lux), level II (320-325 lux), level III (650-655 lux), level IV (1020-1140 lux), 80 degree side light (400-405 lux), and behind light (320-325 lux). The results of the statistical analysis indicated that illumination levels and illumination direction did not affect the performance of the 3D face recognition algorithm, when one enrollment at an illumination level of 220-225 lux was used. In addition, the statistical analysis revealed that the back/behind light similarity scores were closer to those using the full-frontal light similarity scores than the side light. 

Publications

	Kukula*, E. P., & Elliott, S. J. (2004, September). Evaluation of a facial recognition algorithm across three illumination conditions. IEEE Aerospace and Electronic Systems Magazine, (19)9.
	Kukula*, E. P., & Elliott, S. J. (2004, September). Effects of illumination changes on the performance of Geometrix FaceVision 3D FRS, Proceedings of the 38th Annual International Carnahan Conference on Security Technology (ICCST) (pp. 331-337). Albuquerque , NM .
	Elliott, S. J., Kukula, E. P., Sickler, N. C. (2004). The challenges of environment and the human biometric device internation on biometric system performance . International Workshop on Biometric Technologies - Special forum on Modeling and Simulation in Biometric Technology, Calgary, Alberta, Canada
	Kukula, E. P.,(2004, September). Effects of Light Direction on the Performance of Geometrix FaceVision 3D Face Recognition System. Paper presented at the Biometric Consortium, Arlington, VA.
	Kukula*, E. P., & Elliott, S. J. (2003, October). Securing a Restricted Site: Biometric authentication at an entry point. Paper presented at the 2003 IEEE International Carnahan Conference on Security Technology (ICCST), Taipei, Tiawan, ROC
	Morton*, J. M., Portell, C. M., Elliott, S. J., & Kukula, E. P., (2003, October). Facial Recognition at Purdue University Airport 2003-2008. Paper presented at the 2003 IEEE International Carnahan Conference on Security Technology (ICCST), Taipei, Tiawan, ROC
	Kukula*, E. P., & Elliott, S. J. (2003, September). Securing a Restricted Site: Biometric authentication at an entry point. Biometric Symposium, Biometric Consortium 2003 Proceedings, Washington D.C.