The publication of image data captured by ubiquitous surveillance devices, such as traffic cameras and security surveillance cameras, would greatly benefit various communities and enable many applications. However, sharing image data with untrusted parties would raise privacy concern due to potential sensitive content, like identities and activities that may be in the images. Standard image obfuscation techniques, such as pixelation and blurring, do not provide effective privacy preservation for people or objects represented in the data. The goal of this project is to quantitatively define the notion of privacy in image data and develop image publication solutions to achieve rigorous privacy guarantees. By formally modeling image privacy, this project promises significant impact in enabling image data sharing with a wide range of recipients while ensuring individual privacy.
This project develops rigorous privacy notions based on differential privacy for image data, while accounting for the representation of sensitive content in images and effective image obfuscation algorithms to guarantee privacy, and incorporates widely adopted feature extraction techniques in computer vision. The investigator will evaluate the utility and efficiency of the obfuscation algorithms, including the feasibility for popular image processing applications such as crowd counting and object recognition. The project also includes educational activities for K-12 students and involvement of women and minorities.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
