NIST Releases Study on Face Recognition Software

The National Institute of Standards and Technology, NIST, recently released the results of a study to examine the effects of race, age, and sex on face recognition software. A number of contributors are identified in the study as having an effect on the system’s ability to match two images as being from the same person.

Contributors identified in the study include

  • The algorithm at the heart of the system
  • The application the system uses
  • The dataset used
The need for a representative and balanced datasets and the effect they have on results is explored in Women in Standards article “Gender Inclusion in Technology” found here.
Image of facial recognition with quote from article

The NIST study found that “the majority of face recognition algorithms exhibit demographic differentials.” Meaning, the ability of the system to match two images from the same individual was dependent on the demographics of the individual.

The study is unique in that it’s the first to consider the algorithm’s performance as it relates to the demographics of the subject. To evaluate the algorithms, NIST used 18.27 million images, each with metadata identifying information on the subject such as age, sex, and either race or country of birth. Experts say bias in these algorithms can be reduced by using training data sets that are more diverse. The findings provide support for some of the mounting concerns that even the most advanced facial algorithms may not be ready to be used in national security or law enforcement.

To review the study and background on its development, click here.

Continue your reading...

Want to be a part of change? Join us!

Reach out to us today and get a complimentary business review and consultation.