Securics Logo
Securics: The science of securityTM
The Securics Team Biometrics Technology Biometrics Tools Partners Links Contact
Securics currently has products in development in three primary areas: Securics' technology is based on over 5 years of significant funding to Dr. Boult's university research, plus 2 years of extensions and development at Securics. Securics has been awarded multiple SBIR and STTR's focused advancing and commercializing its technology. Securics does not release white-papers. Some of our DOD related work restricted from publications. The rest of our core technologies and performance is published in peer reviewed venues. Copies upon on request. The following papers describe some of the core technologies underlying Securics' developments.
  • W. Scheirer and T. Boult, "Finger Biotope(tm), Overview and Security Analysis", IEEE Computer Vision and Pattern Recognition, June 2007.
  • Binglong Xie, Visvanathan Ramesh, Ying Zhu Terry Boult "On Channel Reliability Measure Training for Multi-Camera Face Recognition" IEEE Workshop on the Application of Computer Vision, Feb 2007.
  • B. Xie, T. Boult, V. Ramesh, Y. Zhu, "Multi-Camera Race Recognition by Reliability-Based Selection", IEEE Conference on Computational Intelligence for Homeland Security and Personal Safety, October 2006.
  • T. E. Boult, "Robust distance measures for face recognition supporting revocable biometric tokens", IEEE Int. Conf. on Face and Gesture, April 2006.
  • T. E. Boult, "PICO: Privacy through Invertible Cryptographic Obscuration", IEEE/NSF Workshop on Computer Vision for Interactive and Intelligent Environments, Nov 11, 2005.
  • T.P. Ripokia and T.E. Boult, "Classification Enhancement via Biometric Pattern Perturbation". IAPR Conference on Audio- and Video-based Biometric Person Authentication, (AVBPA Springer Lecture Notes in Computer Science 3546) pp850-859, July 2005.
  • W. Li, X. Gao, Y. Zhu, V. Ramesh & T.E. Boult, "On the Small Sample Performance of Boosted Classifiers" IEEE Conf. on Computer Vision and Pattern Recognition, Page(s):574 - 581 vol. 2., June 2005
  • T.E. Boult, "Ultra-wide field of view face recognition", Biometric Symposium Sept. 2005.
  • G. Zheng, T.E. Boult,C.-J. Wang, "Projective Invariant Hand Geometry: An overview", Biometric Symposium, Sept. 2005.
  • W. Li, X. Gao and T.E. Boult, "Predicting Biometric System Failure", IEEE Conference on Computational Intelligence for Homeland Security and Personal Safety, March 2005.
  • G. Zheng C.J. Wang and T.E.Boult ``Personal Identification by Cross Ratios of Finger Features, IAPR workshop on Biometrics Challenges from Theory to Practice, August. 2004

Privacy: BiotopesTM which are Secure Robust Revocable Biometric tokens

As biometric begin to enjoy wider deployment, an area of growing concern privacy. A 2002 poll Commissioned by SEARCH, a group funded by the U.S. Bureau of Justice Statistics, looked at perception among biometrics users. Of those who underwent some kind of biometric identification in 2002, the survey found that 88 percent were concerned about possible misuse of their personal information. If existing users feel this way, one must ask how many organization companies that might deploy biometrics are put off, or delaying, because of such privacy concerns.

The concern is more than a simple privacy issue, because if a traditional biometric signature is compromised or stolen, it cannot be replace. Credit Cards, even social security cards, can be replaced if they are stolen, but traditional biometrics are permanent, yet can be duplicated and used to spoof many existing sensors.

Securics is developing products providing BiotopesTM, which are provably secure revocable biometric-based tokens. That is i.e. biometric-based tokens that retain the users privacy yet provide for verification and possibly identification. If compromised they can be revoked and a new one issued. Furthermore different ones can be issued for different applications, prohibiting their use in linking databases thereby reducing privacy concerns. These resulting Biotopes are uniquely matchable but cannot be inverted to obtain the original biometrics template (let alone the raw biometric data). This technology be used with almost any biometric and has a natural formulation for multi-factor security which can be used for verification but not identification. This approach requires minimal added computation for verification -- it could be computed within a smart-card.

Biotope generation process
(Click on image for larger view).		 The FingerBiotope process starts from an image from which it extract minutiae (or it can start from a standard minutiae template). These are show in yellow as there is still a potential for data compromise (if not in a protected hardware) at these stages, which is why Securics is currently working with embedded or encrypting sensors, e.e. so raw fingerprint data does not move over the USB bus. From the template data, Securics generates a Biotope with various keys. It can then transform that Biotope again, if desired, using transactional keys. The "stored" biotope can be looked up in the DB (and transformed with the transactional key if desired). The stored and live Biotopes are then matched in encoded space (i.e. no decrypting) to determine if there is potential match.

For our FingerBiotope, we have demonstrated at CVPR, ICCV and the BCC systems using just a PC, using smart-cards and using embedded processors. To help visualize the revoking process we have a visualization of the Biotopes. Biotopes have multiple keys, and changing any one key produces radical changes in the Biotope representation. The image on the right shows this visually. Note how the same key for different fingers/sensors produces results that are closer than the same image with different keys.

Unlike other work attempting to produce privacy preserving biometrics, the approach developed by Dr. Boult, supports a range of robust distance metrics needed for approximated matching and has been shown so improve performance of existing algorithms. Thus the Securics patent pending approach supports the unique ability of approximate matching while still in "encoded" form. For face recognition and fingerprints the combination of the robust distance measure and unique embedding in higher-dimensional spaces has been shown to significantly improve recognition and verification! For more detail contact us to get our published papers on the topic or to discuss how we might partner.

Biotope matrix for multiple keys
(Click on image for larger view).

The Biotope family of solutions is being developed into two initial product lines, PrivateWhoTM and PrivateMeTM.

PrivateWhoTM supports deployment specific recognition while maintaining the revocable and privacy preserving nature of the underlying technology. This product will allow "recognition" (i.e. answer who is it) within the enrolled population, but the revocable templates cannot be used to link individuals across different deployments or applications. It support detection of multiple enrollments by the same individual (e.g. fraudulent attempts to obtain multiple passport or driver licenses). PrivateWhoTM is ideally suited for applications such as electronic passports, limited enrollment government program (e.g. food stamps), driver licenses, etc. PrivateWhoTM will also support biometric-based login, recognizing the individual from the biometric-derived data, while supporting revocation of that biometric, including an optional centralized credentialing or federated ID service.

PrivateMeTM which supports individual authentication but cannot be used for identification, recognition or search. This technique combines multiple factors into a single secure signature that supports individuals proving "its really me" without revealing the underlying biometric data or giving up their privacy. The PrivateMeTM signature can be used across multiple applications (including an optional centralized credentialing or federated ID service) or can the application can require each user have a specific PrivateMeTM signature that cannot be used for any other application. With end-user controlled revocation and no search abilities, PrivateMeTM is ideal for "on-line verification" from end-user or public locations.

Securics is seeking strategic partners wishing to extend their biometric software/sensors to provide secure revocable biometrics. Securics can help you extend your customer base to those who would not adopt biometrics because of privacy concerns.

 

Biometric Verification and Fusion

Biometric systems depend on similarity measurements as the primary component of the identification/recognition process. Recent developments by Dr. Boult and students have lead to the development of a general theory, Feature Analysis using Similarity Surface Theory (FASST), which can be used to verify the output of a recognizer (or identifier) which in turn can be used as a key component of a fusion module. Securics is extending these developments to address two major biometrics issues: improving system performance (especially for for uncooperative subjects) and developing a theory/system for effective biometric fusion. Our approach simultaneously addresses both problems in a fusion paradigm. Rather than ad-hoc fusion approaches, which have already been tried in multiple academic labs, we seek a unified approach to both individual biometric system improvement and biometric fusion. In both cases the primary component of the approach is taking a "systems view" and building on a theory that allows the system to predict when a particular biometric matching is likely to "fail". For multi-modal fusion, this provides a way to take multiple measurements and weight them based on likelihood of failure. For uncooperative subjects, this failure prediction allows considerations of samples from multiple cameras and/or time, and selection of data are most likely to produce successful results. If combined with other sensor systems, this has significant potential to improve overall performance for recognition of non-cooperating subjects. The approach is based on a patent pending theory for predicting failure of biometric systems. We have been exploring two different algorithms to implement FASST-- one based on Ada-Boosting and one based on multi-layer neural nets using wavelet features. These build on patent pending work by Dr.Boult was refereed to Post-Recognition Analysis Technique (PRAT), because it uses data available after the initial attempt at recognition.
Performance of biometric verification/failure prediction Figure 1: The ROC curves show prediction FA/MD rates for both the training and test sets for images collected under outdoor weather conditions using NTSC video cameras approximately 100ft and 200ft. The distance and weather makes face-based recognition more difficult, with an overall recognition rate of approximately 75% for FaceIt , a commercial product from Identix. The training/testing for this used 21,535 images (split equally). The ROC curve allows one to understand the tradeoffs in performance. For example the third "test" point from the left show that with a FA rate around 3%, the system still correctly predicts more than 75% of the recognition system failures. If a false alarm means taking/processing another image from the video sequence, as for monitoring uncooperative subjects at a distance, a higher FA rate would be acceptable, and increasing the FA rate to 30%, allows a prediction rate around 95%. The potential for using this type of information for fusion (either over time or modalities) should be clear. Depending on the weighting in a fusion approach, even a moderate false-alarm rate is acceptable as it means we are simply depending on the other modality/data more frequently, which is better than mixing incorrect "results" into the fusion algorithm.

The second approach, uses a back-propagation neural-network with wavelet features from the similarity surface. This particular approach was developed to address a particular way of improving a biometric system, using failure prediction to predict when a system was likely to fail, and if so, perturbing key parameters to obtain better results. The "perturbations" considered were eye locations (see Figure 2). Note that this example is a special case of biometric fusion, where the fusion is within the same modality, and is particularly well suited to uncooperative subjects. The algorithm was tested on the weather face datasets taken at 100 and 200 ft, and using two different algorithms, FaceIT, one of the leading the commercial product. Figure 2 shows the types of eye perturbations considered and overall improvement combining FaceIT with the FASST perturbations. Four different times of day throughout the month of May were used for this analysis. The results are shown averaged over multiple days, grouped by time of day. The error bars show 95% statistical confidence. It is clear that the improvements were statically significant.

Figure 2: Perturbations combined with prediction of failure. When the system predicts failure it tries perturbed eye locations. The approximate "perturbations" considered overlaid on the correct eye positions is shown below. The graph on the right shows the base performance and improvements with prediction and perturbations. Testing was on a FaceIt using weathered face data taken at approximately 100ft.

Perturbed eye positions

Biometric performance improvements using pertubations

Hand-geometry: Camera-based Projective invariant hand-geometry recognition

After introducing the above products, Securics will be developing a a touch-free non-intrusive technique of extracting personal hand geometry biometric data. This patent pending technique is based on the invariant property of the planar features on the hand. Unlike previous work, the technique is projective invariant meaning that it does not depend on the orientation of the hand, greatly reducing the constraints on deployment.

This novel approach does not require contact, no pegs and the only constraint is that the hand must be generally stretched flat as in these examples The approach can be applied with images from conventional NTSC or digital still cameras.

Because of the minimal constraints and the use of standard cameras this technology is ideal for fusion with face-based biometrics, where the end-users can just wave at the camera and the system can get a multi-modal biometric which may help with problems of lighting, glare, blinking or other potential failures of the face-based biometric.

4 example hand images

face/hand image

 

Securics is building on technology developed by Dr. T. Boult at the University of Colorado at Colorado Springs and the Colorado Institute of Technology Transfer and Implementation (The projective invariant technique is work conjunction with C. Wang in UCCSs ECE department and G. Zhang a doctoral students in ECE).

Identix and FaceIT are trademarks of Identix Corporation . The work was conducted using generation 4 of the the Identix Face IT SDK, without their direct involvement. The G4 product used for this work is not their current release version.

-->

Copyright 2004, 2005, T. Boult      Valid HTML 4.01!