Blur vs. Block: Investigating the Effectiveness of Privacy-Enhancing Obfuscation for Images

This paper aims to improve privacy-preserving visual recognition, an increasingly demanded feature in smart camera applications, by formulating a unique adversarial training framework. The proposed framework explicitly learns a degradation transform for the original video inputs, in order to optimize the trade-off between target task performance and the associated privacy budgets on the degraded video. A notable challenge is that the privacy budget, often defined and measured in task-driven contexts, cannot be reliably indicated using any single model performance, because a strong protection of privacy has to sustain against any possible model that tries to hack privacy information. Such an uncommon situation has motivated us to propose two strategies, i.e., budget model restarting and ensemble, to enhance the generalization of the learned degradation on protecting privacy against unseen hacker models. Novel training strategies, evaluation protocols, and result visualization methods have been designed accordingly. Two experiments on privacypreserving action recognition, with privacy budgets defined in various ways, manifest the compelling effectiveness of the proposed framework in simultaneously maintaining high target task (action recognition) performance while suppressing the privacy breach risk. The code is available at https://github.com/wuzhenyusjtu/Privacy-AdversarialLearning

show abstract

“…We compare the following five methods: -Method 1 (naive downsampling): using raw RGB frames under different down-sampling rates . [26].…”

Section: Identity-preserving Action Recognition On Sbumentioning

confidence: 99%

Towards Privacy-Preserving Visual Recognition via Adversarial Training: A Pilot Study

Wang

et al. 2018

Lecture Notes in Computer Science

132

View full text Add to dashboard Cite

show abstract

“…Most of these approaches only consider about the privacy leakage problems, but offer no guarantee in the scenario where users still need to enjoy content sharing in OSN. As shown in [38], the privacy-enhancing technologies should provide a trade-off between the privacy and a good viewing experience. In our work, we proposed to adopt adversarial examples as the potential solution.…”

Section: Related Workmentioning

confidence: 99%

Recessive Social Networking: Preventing Privacy Leakage against Reverse Image Search

Zhang

Lin³

2019

2019 IEEE European Symposium on Security and Privacy Workshops (EuroS&PW)

View full text Add to dashboard Cite

This work investigates the image privacy problem in the context of social networking under the threat of reverse image search. We introduce a new concept called recessive social networking. Unlike conventional privacy-preserving social networking, in our setting, the aim is to deceive machine learning algorithms that used in reverse image search, while still enabling unaffected ubiquitous social networking among humans. We, for the first time, ultilize adversarial example technique as a defensive mechanism to protect image privacy against content-based image search algorithms in the context of social networking. Finally, rigorous evaluations are conducted to demonstrate the effectiveness, transferability, and robustness of the proposed countermeasure.

show abstract

“…Although the blooming development has advanced machine learning applications to bring convenience and enhance user experience, it may capture confidential information incidentally and increases the risks of privacy leakage. To protect privacy, the most straightforward approach is access control [51], such as the "Restrict Others" setting in Facebook [31]. Cryptography techniques such as encryption and secure multi-party computation [28] can also be applied to mitigate Figure 1: A brief comparison of obfuscation methods.…”

Section: Introductionmentioning

confidence: 99%

“…Second, images processed with these techniques result in unsatisfying perception in general. A study shows that both blurring and blocking will impact image perception scores, and even lower scores are observed for images obfuscated by blocking [31]. On the other hand, new techniques and mechanisms are being applied to enhance image obfuscation.…”

Section: Introductionmentioning

confidence: 99%

AnonymousNet: Natural Face De-Identification With Measurable Privacy

Lin

2019

2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

133

View full text Add to dashboard Cite

show abstract

Blur vs. Block: Investigating the Effectiveness of Privacy-Enhancing Obfuscation for Images

Cited by 47 publications

References 55 publications

Towards Privacy-Preserving Visual Recognition via Adversarial Training: A Pilot Study

Towards Privacy-Preserving Visual Recognition via Adversarial Training: A Pilot Study

Recessive Social Networking: Preventing Privacy Leakage against Reverse Image Search

AnonymousNet: Natural Face De-Identification With Measurable Privacy

Contact Info

Product

Resources

About