Model-Based Face De-Identification

By Ralph Gross, Latanya Sweeney, Fernando De la Torre, Simon Baker. 2006

221 citations recorded by the source corpus.

Advances in camera and computing equipment hardware in recent years have made it increasingly simple to capture and store extensive amounts of video data. This, among other things, creates ample opportunities for the sharing of video sequences. In order to protect the privacy of subjects visible in the scene, automated methods to de-identify the images, particularly the face region, are necessary. So far the majority of privacy protection schemes currently used in practice rely on ad-hoc methods such as pixelation or blurring of the face. In this paper we show in extensive experiments that pixelation and blurring offers very poor privacy protection while significantly distorting the data. We then introduce a novel framework for de-identifying facial images. Our algorithm combines a model-based face image parameterization with a formal privacy protection model. In experiments on two large-scale data sets we demonstrate privacy protection and preservation of data utility.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Face swapping: automatically replacing faces in photographs (2008, methodology): “The work of [Gross et al. 2006] improves the quality of de-identiﬁed faces using Active Appearance Models [Cootes et al. 2001] but still suffers from blurring artifacts.”
DeepPrivacy: A Generative Adversarial Network for Face Anonymization (2019, background): “K-same family of algorithms [4,11,20] implements the k-anonymity algorithm [25] for face images.”
Getting the Face Behind the Squares: Reconstructing Pixelized Video Streams (2011, background): “show that pixelization has very low effectiveness in preserving privacy when pixelized video sequence is fed to a face recognition software matching the face against a face database [8]; quite unexpectedly, they also find that attempts to reconstruct the original image usually worsen the effectivene”
Semi-adversarial Networks: Convolutional Autoencoders for Imparting Privacy to Face Images (2017, background): “On one hand, there are face de-identification techniques [11, 20, 7] where a face image is modified in order to confound a face matcher.”