DP-Image: Differential Privacy for Image Data in Feature Space

Liu, Bo; Ding, Ming; Xue, Hanyu; Zhu, Tianqing; Ye, Dayong; Song, Li; Zhou, Wanlei

doi:10.48550/arxiv.2103.07073

Cited by 4 publications

(21 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…n Z [21,22,5,46]. We demonstrate that to ensure -LDP, the d Z -maximization property of the identityobfuscation function has to be relaxed.…”

Section: Formalismmentioning

confidence: 99%

“…Moreover, the two domains consider different performance indicators and evaluation metrics. Anonymization aims at providing privacy preserving guarantees, including face anonymization rate and non re-identifiability [26,46,15,66], which implies additional mechanisms compared to the face-swapping methods that prioritize preserving facial attributes while reckoning the visual quality of the injected identity [55,77]. Face Anonymization.…”

Section: Related Workmentioning

confidence: 99%

“…A significant amount of research on face anonymization approaches the problem as an image inpainting task, where the face region is first erased and then replaced with another. Early methods [26,33,56] use a database of real faces to aggregate the new identity, while more recent methods [32,50,46] use generative models to synthesize fake identities based on the learned distribution. DeepPrivacy [32] is one of the pioneering works in this field, which reconstructs the missing face by taking the masked face and facial landmarks as inputs.…”

Section: Related Workmentioning

confidence: 99%

“…In this work, we propose to anonymize faces in an irreversible manner. Other solutions [45,13,44,46] incorporate notions of differential privacy [21,22,5] by adding adequately-calibrated random noise either at training or inference time, ensuring privacy levels linked to their parameter . However, they often neglect utility preservation (e.g., they edit image background and utility attributes) and require complex post-processing , making them not readily applicable to anonymization tasks.…”

Section: Related Workmentioning

confidence: 99%

“…Identity Obfuscation Guarantees. To provide formal de-identification guarantees, we ground our work in the extensive theory on -differential privacy ( -DP) andlocal-differential privacy ( -LDP, relevant when obfuscation should be performed without global knowledge) applied to identity-swapping functions [21,22,5,79,46,15,66,59]. Let ψ : Z → Z be a function that performs ID obfuscation, i.e., taking an identity vector z and returning a new one z that maximizes d Z (z, z).…”

Section: Formalismmentioning

confidence: 99%

See 4 more Smart Citations

Disguise without Disruption: Utility-Preserving Face De-Identification

Cai¹,

Gao²,

Planche³

et al. 2023

Preprint

View full text Add to dashboard Cite

With the increasing ubiquity of cameras and smart sensors, humanity is generating data at an exponential rate. Access to this trove of information, often covering yetunderrepresented use-cases (e.g., AI in medical settings) could fuel a new generation of deep-learning tools. However, eager data scientists should first provide satisfying guarantees w.r.t. the privacy of individuals present in these untapped datasets. This is especially important for images or videos depicting faces, as their biometric information is the target of most identification methods. While a variety of solutions have been proposed to de-identify such images, they often corrupt other non-identifying facial attributes that would be relevant for downstream tasks.In this paper, we propose Disguise, a novel algorithm to seamlessly de-identify facial images while ensuring the usability of the altered data. Unlike prior arts, we ground our solution in both differential privacy and ensemble-learning research domains. Our method extracts and swaps depicted identities with fake ones, synthesized via variational mechanisms to maximize obfuscation and non-invertibility; while leveraging the supervision from a mixture-of-experts to disentangle and preserve other utility attributes. We extensively evaluate our method on multiple datasets, demonstrating higher de-identification rate and superior consistency than prior art w.r.t. various downstream tasks.

show abstract

“…n Z [21,22,5,46]. We demonstrate that to ensure -LDP, the d Z -maximization property of the identityobfuscation function has to be relaxed.…”

Section: Formalismmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Formalismmentioning

confidence: 99%

See 3 more Smart Citations

Disguise without Disruption: Utility-Preserving Face De-Identification

Cai¹,

Gao²,

Planche³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

Disguise without Disruption: Utility-Preserving Face De-identification

Cai,

Gao,

Planche

et al. 2024

AAAI

View full text Add to dashboard Cite

With the rise of cameras and smart sensors, humanity generates an exponential amount of data. This valuable information, including underrepresented cases like AI in medical settings, can fuel new deep-learning tools. However, data scientists must prioritize ensuring privacy for individuals in these untapped datasets, especially for images or videos with faces, which are prime targets for identification methods. Proposed solutions to de-identify such images often compromise non-identifying facial attributes relevant to downstream tasks. In this paper, we introduce Disguise, a novel algorithm that seamlessly de-identifies facial images while ensuring the usability of the modified data. Unlike previous approaches, our solution is firmly grounded in the domains of differential privacy and ensemble-learning research. Our method involves extracting and substituting depicted identities with synthetic ones, generated using variational mechanisms to maximize obfuscation and non-invertibility. Additionally, we leverage supervision from a mixture-of-experts to disentangle and preserve other utility attributes. We extensively evaluate our method using multiple datasets, demonstrating a higher de-identification rate and superior consistency compared to prior approaches in various downstream tasks.

show abstract

Local Differential Privacy Image Generation Using Flow-Based Deep Generative Models

Shibata,

Hanaoka,

Cao

et al. 2023

Applied Sciences

View full text Add to dashboard Cite

Diagnostic radiologists need artificial intelligence (AI) for medical imaging, but access to medical images required for training in AI has become increasingly restrictive. To release and use medical images, we need an algorithm that can simultaneously protect privacy and preserve pathologies in medical images. To address this, we introduce DP-GLOW, a hybrid that combines the local differential privacy (LDP) algorithm with GLOW, one of the flow-based deep generative models. By applying a GLOW model, we disentangle the pixelwise correlation of images, which makes it difficult to protect privacy with straightforward LDP algorithms for images. Specifically, we map images to the latent vector of the GLOW model, where each element follows an independent normal distribution. We then apply the Laplace mechanism to this latent vector to achieve ϵ-LDP, which is one of the LDP algorithms. Moreover, we applied DP-GLOW to chest X-ray images to generate LDP images while preserving pathologies. The ϵ-LDP-processed chest X-ray images obtained with DP-GLOW indicate that we have obtained a powerful tool for releasing and using medical images for training AI.

show abstract

DP-Image: Differential Privacy for Image Data in Feature Space

Cited by 4 publications

References 46 publications

Disguise without Disruption: Utility-Preserving Face De-Identification

Disguise without Disruption: Utility-Preserving Face De-Identification

Disguise without Disruption: Utility-Preserving Face De-identification

Local Differential Privacy Image Generation Using Flow-Based Deep Generative Models

Contact Info

Product

Resources

About