To understand how efficient Aureus 3D face recognition software is at finding and matching faces we feel it’s important for you to have access to the best information possible. To that end, results from well-known and publicly-available data sets are shown below. If you would like to independently verify our results, we’ll provide you with the Aureus 3D SDK and guidance to get up and running.
The foundation of your face recognition solution is found within Aureus 3D.
The National Institute of Standards and Technology (NIST) is respected around the world as an independent evaluation and testing body. For more than two decades NIST has been investigating the performance of different biometric technologies. The various iterations of the Facial Recognition Vendor’s Test (FRVT) have been globally lauded as producing the definitive testing protocols.
In NIST’s FRVT 2002 report, NIST affirmed CyberExtruder’s approach to face recognition by stating “FRVT 2002 also assessed the impact of three new techniques for improving face recognition: three-dimensional morphable models, normalization of similarity scores, and face recognition from video sequences. Results show that three-dimensional morphable models and normalization increase performance.”
In NIST’s FRVT 2013 Class F (Frontalization) testing, NIST compared the performance of algorithms from Cogent (A20C), Morpho (D20C), NEC (E20C) and Toshiba (J20C) and concluded that CyberExtruder’s Aureus 3D Frontalization approach produced results demonstrably better than the competition’s conventional approach.
In the NIST Face in Video Evaluation (FIVE), CyberExtruder’s Aureus 3D version 5.3 algorithm was tested and results are included in the 2017 NIST report. We are very proud of our performance in the first ever comparison of algorithms on video datasets.
Independent Testing and Evaluations
Our approach to face recognition has evolved as computing power has increased and market requirements have sharpened. Along the way, we have either asked to be tested or been approached by respected individuals interested in evaluating our unique approach. We always encourage outside scrutiny and feedback as it provides a necessary step on the road to improving our technology.
Current – We are participating in NIST’s Ongoing FRVT evaluation
Current – We are working with Dr. Michael King of the Harris Institute for Assured Information at Florida Institute of Technology
2014 – Lacey Best-Rowden and Dr. Anil Jain at Michigan State University evaluated Aureus 3D as part of their publication “Unconstrained Face Recognition: Identifying a Person of Interest from a Media Collection”
2011 – Rutger Storm at Unisys, and in cooperation with the Utrecht University of Applied Sciences, Netherlands, evaluated Aureus 3D as part of the research for his publication “Impact of image morphing on face recognition”
2004- Dr. Ernst Mucke conducted an investigation for Identix which used Aureus 3D to develop a first version of an operational surveillance system using 3D face models and Identix’ ABIS 3.0 search engine.
They say we are our own worst critic, and in our business, we have to be. We challenge ourselves to be better every day, and we continuously reinvest in our technology. Our current algorithms (Aureus 3D version 5.7) reflect that drive for excellence.
To illustrate our current level of performance we’re presenting results derived from NIST’s FERET and FRGC data sets, the University of Massachusetts’ Computer Vision Lab’s Labeled Faces in the Wild (LFW) data set, and our in-house Ultimate data set.
FERET Data set
NIST’s FERET data set is a great collection of 7,833 facial images of 955 people displaying examples of extreme pose which range from near frontal to nearly profile.
Aureus 3D 5.7 Performance on the FERET Data set
LFW Data set
The Labeled Faces in the Wild data set contains 9,125 facial images of 1,673 people designed for studying the problem of unconstrained face recognition. This data set contains more than 13,000 images of faces collected from the Internet.
Aureus 3D 5.7 Performance on the Labeled Faces in the Wild Data set
FRGC Data set
The FRGC set was created by NIST for the 2010 Face Recognition Grand Challenge. It contains 22,549 images of 466 people. Variable lighting conditions across the face are the primary distinguishing factor of this set.
Aureus 3D 5.7 Performance on the FRGC Data set
Ultimate Data set
The Ultimate data set is our own internal collection of images which we feel represent the hardest test bed possible. It includes extreme instances of pose, lighting, age differences, low resolution images and even comparisons of artist’s sketches. It contains 10,210 examples of 1,000 people.
Aureus 3D 5.7 Performance on our Ultimate Data set