Facial recognition systems are used around the world to check very quickly if an individual's photo appears in a database of known or suspected criminals. Unfortunately, the systems often acquire poor facial images in real-world environments such as airports. Now, researchers at the U.S. National Institute of Standards and Technology (NIST) have found that several simple steps can significantly improve the quality of facial images that are acquired at border entry points such as airports and seaports. The researchers add that their recommendations for improving facial images could be implemented relatively easily -- and cheaply -- with existing facial recognition technology. Read more...
You can see above how the NIST facial recognition system rates a number of face attributes. These attributes are described by two separate ANSI and ISO norms. "Since there are no available fully automated conformance tests based on these standards, the approach used to evaluate the quality of the images was visual inspection to rate the first 14 attributes on a five-point scale where a value of one is least conducive to automatic face recognition and five is most conducive. The remaining six attributes are binary and were assigned values of 1 or 5, where 1 indicated the presence of the characteristic and 5 the absence."
For your information, the two international norms mentioned above are ANSI-INCITS 385-2004 (Face Recognition Format for Data Interchange) and ISO/IEC 19794-5:2005(E) (Biometric Data Interchange Formats - Part 5: Face image data).
This new research work has been led by Mary Theofanos and her colleagues of the Visualization and Usability Group (VUG) at the U.S. National Institute of Standards and Technology. These scientists worked with members of the US-VISIT program implemented by the Department of Homeland Security (DHS).
Now, let's see how the two teams of researchers identified a number of steps to take for acquiring better facial images. Their report "recommends that operators should adjust camera settings to ensure the subject comes into sharp focus. The report also recommends using a traditional-looking camera in facial-recognition systems so that individuals could clearly recognize the camera and look into it. Following the Dulles site visit, a study adopted these steps in taking facial images of 300 participants while mimicking the real-world conditions of a border entry point."
This research work has been published as a NIST Interagency Report in September 2008 under the name "Assessing Face Acquisition" (PDF format, 39 pages, 948 KB). The above picture has been extracted from this document. Here are five usability and human factors enhancements to the current US-VISIT collection process provided by this report.
- The camera should resemble a traditional camera.
- The camera should click when the picture is taken to provide feedback to the traveler that the picture is being taken.
- The camera should be used in portrait mode.
- The operator should be facing the traveler and the monitor while positioning the camera and provide some marking on the floor (such as footprints) to indicate to the traveler where to stand for the photograph.
And here are the three main results from this report. "The enhancements were designed to address the extreme conditions or departures in the captured images. Implementing these enhancements resulted in:
- 100% of the images capturing a participant's face in contrast to the current US-VISIT collection
- All of the participants were facing the camera -- this is a significant improvement to the process currently used at the ports of entry.
- Additional improvement may be realized by using the face overlay guide proactively. By incorporating the overlay into the workstations the officers could use the guide to center the camera on the participant's face.
According to the report, "these recommended enhancements improved the overall captured images and can be implemented relatively easily and with relatively little cost. A follow-up study incorporating the overlay into the operators' workflow is underway."
Sources: National Institute of Standards and Technology (NIST), November 25, 2008; and various websites
You'll find related stories by following the links below.