Vision: Face de-identification
Problem Statement: Given video or a photograph, de-identify faces appearing in the video or photograph so that no matter how good face recognition software may become, it cannot reliably recognize the faces yet facial details remain in the image.
Description: The k-Same algorithm by Dr. Sweeney and students is a solution. It scientifically limits the ability of face recognition software to reliably recognize faces while maintaining facial details in the images. The algorithm determines similarity between faces based on a distance metric and creates new faces by averaging image components, which may be the original image pixels (k-Same-Pixel) or eigenvectors (k-Same-Eigen). Results are presented on a standard collection of real face images with varying k. We also show how ad hoc techniques (e.g., pixelation and additive noise) do not work. Later papers describe methods for producing photo realistic images in real-time using active appearance models and multi-factor models. Its privacy guarantee limits face recognition to do no better than guessing 1/k.
Scientific Influence and Impact: Dr. Sweeney and her students were first to demonstrate the importance of using provable privacy protection over ad hoc approaches, by showing how face recognition could re-identify faces distorted by masking, additive noise, and pixelation. They introduced the first formal model for protection. Others introduced alternatives and enhancements [Defaux et al.]. Senior recenty edited a book on the topic [cite]. Working with Gross, Cohn, de la Torre and Baker, they produced anonymized, photo realistic video of pain grimace in patients for NIH [cite].
Other Achievements: 12