mlsanityData to 14 Oct 2026

Analysis and Mitigations of Reverse Engineering Attacks on Local Feature Descriptors

arXiv 2105.03812 ↗First posted 9 May 2021cs.CVcited by 10

Not on the map. Why?

Abstract (arXiv)

As autonomous driving and augmented reality evolve, a practical concern is data privacy. In particular, these applications rely on localization based on user images. The widely adopted technology uses local feature descriptors, which are derived from the images and it was long thought that they could not be reverted back. However, recent work has demonstrated that under certain conditions reverse engineering attacks are possible and allow an adversary to reconstruct RGB images. This poses a potential risk to user privacy. We take this a step further and model potential adversaries using a privacy threat model. Subsequently, we show under controlled conditions a reverse engineering attack on sparse feature maps and analyze the vulnerability of popular descriptors including FREAK, SIFT and SOSNet. Finally, we evaluate potential mitigation techniques that select a subset of descriptors to carefully balance privacy reconstruction risk while preserving image matching accuracy; our results show that similar accuracy can be obtained when revealing less information.

Key facts. Cited by 10 papers in the map (data through 14 Oct 2026). Builds on 25 of its 57 references that are in the map. First posted on arXiv 9 May 2021.

Builds on

25 of 57 references.

References · 25

  1. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  2. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,713
  3. 2013Intriguing properties of neural networkscited by 5,155
  4. 2018YOLOv3: An Incremental Improvementcited by 2,659
  5. 2015Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networkscited by 2,054
  6. 2014Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Imagescited by 776
  7. 2018MegaDepth: Learning Single-View Depth Prediction from Internet Photoscited by 737
  8. 2017Evasion Attacks against Machine Learning at Test Timecited by 689
  9. 2018Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Surveycited by 460
  10. 2014Understanding Deep Image Representations by Inverting Themcited by 403
  11. 2018Gazelle: A Low Latency Framework for Secure Neural Network Inferencecited by 193
  12. 2015Inverting Visual Representations with Convolutional Networkscited by 170
13 more
  1. 2019SOSNet: Second Order Similarity Regularization for Local Descriptor Learningcited by 131
  2. 2019Revealing Scenes by Inverting Structure from Motion Reconstructionscited by 66
  3. 2018Learning to Anonymize Faces for Privacy Preserving Action Detectioncited by 63
  4. 2019AnonymousNet: Natural Face De-Identification with Measurable Privacycited by 50
  5. 2018Towards Privacy-Preserving Visual Recognition via Adversarial Training: A Pilot Studycited by 48
  6. 2019Privacy Preserving Image-Based Localizationcited by 47
  7. 2016Privacy-Preserving Human Activity Recognition from Extreme Low Resolutioncited by 43
  8. 2018Learning Privacy Preserving Encodings through Adversarial Trainingcited by 30
  9. 2019Privacy-Preserving Action Recognition using Coded Aperture Videoscited by 20
  10. 2020Privacy Preserving Visual SLAMcited by 15
  11. 2017Privacy-Preserving Visual Learning Using Doubly Permuted Homomorphic Encryptioncited by 14
  12. 2015Image Reconstruction from Bag-of-Visual-Wordscited by 6
  13. 2020Privacy-Preserving Image Features via Adversarial Affine Subspace Embeddingscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 1; 2022: 1; 2023: 4; 2024: 2; 2025: 1; 2026: 1.

10 papers.

  1. 2021NinjaDesc: Content-Concealing Visual Descriptors via Adversarial Learningcited by 21
  2. 2024Fully Geometric Panoramic Localizationcited by 11
  3. 2023Privacy-Preserving Representations are not Enough -- Recovering Scene Content from Camera Posescited by 7
  4. 2022Privacy-Preserving Visual Localization with Event Camerascited by 6
  5. 2024Obfuscation Based Privacy Preserving Representations are Recoverable Using Neighborhood Informationcited by 6
  6. 2023PS-FedGAN: An Efficient Federated Learning Framework Based on Partially Shared Generative Adversarial Networks For Data Privacycited by 4
  7. 2023LDL: Line Distance Functions for Panoramic Localizationcited by 3
  8. 2023UFed-GAN: A Secure Federated Learning Framework with Constrained Computation and Unlabeled Datacited by 1
  9. 2025A Distributed Framework for Privacy-Enhanced Vision Transformers on the Edgecited by 0
  10. 2026Revisiting Geometric Obfuscation with Dual Convergent Lines for Privacy-Preserving Image Queries in Visual Localizationcited by 0