Face Recognition

Research

Last Updated 12/05/07

Literature Review

Revised Literature Review

Research Proposal

Topic

Facial Recognition for Security Applications

Description

This research reviews techniques for improving the accuracy of facial recognition software for use in security applications.

Motivation

Biometric technology, using physical characteristics to identify humans, is important to the future of security. This technology has potential to replace usernames and passwords as the primary form of security on a computer since it is much more difficult to steal, crack, forget, or otherwise falsify. Facial recognition can also aid in fighting crime. The technology could be used in conjunction with camera footage to track criminals on the run. Improving techniques in this field will help make its applications faster, more accurate, and ultimately more usable.

References

E. P. Vivek, N. Sudha, "Gray Hausdorff Distance Measure for Comparing Face Images," IEEE Transactions on Information Forensics and Security, VOL. 1, NO. 3, 2006.

[The article is about using Hausdorff distances for facial recognition. The method discussed works under the assumption that it is comparing frontal images and that corresponding points in compared images fall within a given neighborhood. The method compares the gray images of faces using Hausdorff distances. The technique is different from previous Hausdorff distance methods because it compares pixel intensity distributions of faces instead of edge maps. A recognition rate of more than 80% has been achieved when testing using select face databases. The technique is found to be fairly robust in changes in pose and expression, and has some robustness with regards to variations in lighting. Techniques for improving the technique's sensativity to light would be a topic for further research.]

Mathhew A. Turk, Alex P. Pentland, "Face Recognition Using Eigenfaces," Proc. IEEE Conference on Computer Vision and Pattern Recognition: 586–591. 1991.

[Face Recognition Using Eigenfaces is one of the most important papers to modern facial recognition research. The technique described is design to be “fast, reasonably simple, and accurate in constrained environments such as an office or household.” The method analyzes face images and computes eigenfaces which are faces composed of eigenvectors. The comparison of eigenfaces is used to identify the presence of a face and its identity. There is a five step process involved with the system developed by Turk and Pentland. First the system needs to be initialized by feeding it a set of training images of faces. It used these to define what a face looks like. Next, when a face is encountered it calculates an eigenface for it. By comparing it with known faces and using some statistical analysis it can be determined whether the image presented is a face at all. Then, if an image is determined to be a face the system will determine whether it knows the identity of it or not. The optional final step is that if an unknown face is seen repeatedly, the system can learn to recognize it. The system was even tested to track faces on film. The technique excels at being computationally efficient and fairly accurate. However, the early techniques used by Turk and Pentland were not especially robust to variations in lighting, orientation of faces, and the size of faces.]

Pawan Sinha, Benjamin Balas, Yuri Ostrovsky, and Richard Russell, "Face Recognition by Humans: Nineteen Results All Computer Vision Researchers Should Know About," Proceedings of the IEEE, Volume: 94, Issue: 11, 2006.

[This paper offers valuable insight into the complexity and robustness of human facial recognition. Nineteen different results are explained and discussed in terms of how it may impact computer facial recognition. The results are as follows:

  1. Humans can recognize familiar faces in very low-resolution images.
  2. The ability to tolerate degradations increases with familiarity.
  3. High-frequency information by itself is insufficient for good face recognition performance.
  4. Facial features are processed holistically.
  5. Of the different facial features, eyebrows are among the most important for recognition.
  6. The important configural relationships appear to be independent across the width and height dimensions.
  7. Face-shape appears to be encoded in a slightly caricatured manner.
  8. Prolonged face viewing can lead to high level aftereffects, which suggest prototype-based encoding.
  9. Pigmentation cues are at least as important as shape cues.
  10. Color cues play a significant role, especially when shape cues are degraded.
  11. Contrast polarity inversion dramatically impairs recognition performance, possibly due to compromised ability to use pigmentation cues.
  12. Illumination changes influence generalization.
  13. View-generalization appears to be mediated by temporal association.
  14. Motion of faces appears to facilitate subsequent recognition.
  15. The visual system starts with a rudimentary preference for face-like patterns.
  16. The visual system progresses from a piecemeal to a holistic strategy over the first several years of life.
  17. The human visual system appears to devote specialized neural resources for face perception.
  18. Latency of responses to faces in inferotemporal (IT) cortex is about 120 ms, suggesting a largely feed forward computation.
  19. Facial identity and expression might be processed by separate systems.

The results open many questions about the processes involved in human face recognition and how they may be mimicked by a computer system.]

Michael Kraus, "Face the facts: facial recognition technology's troubled past--and troubling future," The Free Library, 2002. [Talks about some of the problems the technology has encountered in its applications]

Ryan Johnson, Kevin Bonsor, "How Facial Recognition Systems Work," How Stuff Works, 2007. [A good primer on facial recognition technology and springboard for other areas.]

Andrew Colley, "SmartGate not pulling its weight," ZDNet Australia, February 11, 2004. [Coverage of SmartGate usage in Australia.]

John D. Woodward, Jr., Christopher Horn, Julius Gatune, Aryn Thomas, Biometrics, A Look at Facial Recognition, RAND, 2003.

[Facial recognition is attractive for law enforcement because it can be used in conjunction with cameras. This makes it covert and non intrusive, opposed to other biometrics such as finger prints, retina scans, and iris scans. This is especially important in conjunction with the law because faces are considered public. Face recognition can also use preexisting photo databases from mugshots or driver’s licenses. Because of difficulties face recognition has with respect to lighting, angle, and other factors, it is advantageous to attempt to get as high quality images as possible. Several “facetraps” are proposed, such as placing cameras facing doorways, at airport check-ins, or near objects people are likely to stare at. These traps would aid face recognition software by helping to capture a frontal image which is important to the software. Face recognition must be improved further before it becomes a useful tool for law enforcement. It remains to be seen what the right balance is, legally speaking, between maximizing public safety and respecting individual rights.]

Mark Williams, "Better Face-Recognition Software," Technology Review, May 30, 2007.[Talks about recent enhancements in technology using 3D face recognition and surface texture analysis, really encouraging results.]

Trina D. Russ, Mark W. Koch, Charles Q. Little, "3D Facial Recognition: A Quantitative Analysis," 38th Annual 2004 International Carnahan Conference on Security Technology, 2004.

[ The work presented is a quantitative analysis of 3d facial recognition. Issues with 2d facial recognition are discussed. 2d facial recognition require frontal face views and test have shown significant drops in performance when faces are rotated, large databases are used, or images from different distances are compared. Some complicated issues with 3d face recognition is mentioned. 3d systems need to be robust to changes in expression, hairstyle, and presence of glasses. An algorithm is presented which is robust to these factors. The results showed impressive performance of the system. Future work will involve improving the alignment of images, which should improve computational performance and accuracy. Further tests will involve larger database and images with noise and rotation variation.]