Privacy-Preserving Image Features via Adversarial Affine Subspace Embeddings

Mihai Dusmanu 1, Johannes L. Schönberger 2, Sudipta N. Sinha 2, Marc Pollefeys 1, 2

1ETH Zürich, 2Microsoft

CVPR 2021 (Oral, Best Paper Candidate)

Overview of the proposed method. Inversion of traditional local image features is a privacy concern in many applications. Our proposed approach obfuscates the appearance of the original image by lifting the descriptors to affine subspaces. Distance between the privacy-preserving subspaces enables efficient matching of features. The same concept can be applied to other domains such as face features for biometric authentication. Image credit: laylamoran4battersea (Layla Moran) on Flickr.

Abstract

Many computer vision systems require users to upload image features to the cloud for processing and storage. These features can be exploited to recover sensitive information about the scene or subjects, e.g., by reconstructing the appearance of the original image. To address this privacy concern, we propose a new privacy-preserving feature representation. The core idea of our work is to drop constraints from each feature descriptor by embedding it within an affine subspace containing the original feature as well as adversarial feature samples. Feature matching on the privacy-preserving representation is enabled based on the notion of subspace-to-subspace distance. We experimentally demonstrate the effectiveness of our method and its high practical relevance for the applications of visual localization and mapping as well as face authentication. Compared to the original features, our approach makes it significantly more difficult for an adversary to recover private information.

Video

Code

Coming soon...

Paper

Mihai Dusmanu 1, Johannes L. Schönberger 2, Sudipta N. Sinha 2, Marc Pollefeys 1, 2
1ETH Zürich, 2Microsoft
Privacy-Preserving Image Features via Adversarial Affine Subspace Embeddings
In Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition
[Latest version on arXiv] [Poster] [Video on YouTube]
BibTeX
@InProceedings{Dusmanu2021Privacy,
    author = "Dusmanu, Mihai and Sch\"onberger, Johannes L. and Sinha, Sudipta N. and Pollefeys, Marc",
    title = "{P}rivacy-{P}reserving {I}mage {F}eatures via {A}dversarial {A}ffine {S}ubspace {E}mbeddings",
    booktitle = "Proceedings of the 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition",
    year = "2021"
}