DeepPrivacy: A Generative Adversarial Network for Face Anonymization

Hukkelås, Håkon; Mester, Rudolf; Lindseth, Frank

doi:10.48550/arxiv.1909.04538

Cited by 3 publications

(9 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, we conduct experiments to evaluate the efficacy of our proposed method for facial privacy-protection and -recovery performance compared to existing approaches. Specifically, we compare our method with those proposed by Maximov et al [36], Shan et al [37], Hukkelas et al [38], You et al [39], Yang et al [40], Li et al [41], Gu et al [42] and He et al [43]. Hukkelas et al [38] utilize a GAN-based approach to generate realistic and anonymous images for individual privacy protection.…”

Section: Compare With Other Methodsmentioning

confidence: 99%

“…Specifically, we compare our method with those proposed by Maximov et al [36], Shan et al [37], Hukkelas et al [38], You et al [39], Yang et al [40], Li et al [41], Gu et al [42] and He et al [43]. Hukkelas et al [38] utilize a GAN-based approach to generate realistic and anonymous images for individual privacy protection. Maximov et al [36] employ a GAN-based model to anonymize identity information through conditional generation.…”

Section: Compare With Other Methodsmentioning

confidence: 99%

“…This lower success rate can be attributed to its reliance on adversarial interference, which may not be compatible with the loss functions and training methodologies employed by many commercial face-recognition tools. Conversely, methods proposed by You et al [39], Maximov et al [36], Hukkelas et al [38], Yang et al [40] and our proposed method achieved protection success rates of 100% when tested using the face-recognition tool from Baidu Intelligent Cloud [45]. Notably, You et al's method employs reversible mosaic transformation, rendering the processed faces unrecognizable by Baidu Intelligent Cloud [45].…”

Section: Quantitative Comparison Of Face Privacy-protection Effectmentioning

confidence: 98%

“…In summary, our method effectively safeguards the privacy of protected faces while offering near-perfect recovery of obscured faces. × Hukkelas et al [38] × Utilizing the recovered images depicted in Figure 5, we computed the similarity metrics between the original and recovered images. Specifically, we calculated the PSNR, SSIM, RMSE and MAE values between the protected and recovered surfaces.…”

Section: Quantitative Comparison Of Face Privacy-protection Effectmentioning

confidence: 99%

See 3 more Smart Citations

Generation of Face Privacy-Protected Images Based on the Diffusion Model

You,

Zhao,

Wang

et al. 2024

Entropy

View full text Add to dashboard Cite

In light of growing concerns about the misuse of personal data resulting from the widespread use of artificial intelligence technology, it is necessary to implement robust privacy-protection methods. However, existing methods for protecting facial privacy suffer from issues such as poor visual quality, distortion and limited reusability. To tackle this challenge, we propose a novel approach called Diffusion Models for Face Privacy Protection (DIFP). Our method utilizes a face generator that is conditionally controlled and reality-guided to produce high-resolution encrypted faces that are photorealistic while preserving the naturalness and recoverability of the original facial information. We employ a two-stage training strategy to generate protected faces with guidance on identity and style, followed by an iterative technique for improving latent variables to enhance realism. Additionally, we introduce diffusion model denoising for identity recovery, which facilitates the removal of encryption and restoration of the original face when required. Experimental results demonstrate the effectiveness of our method in qualitative privacy protection, achieving high success rates in evading face-recognition tools and enabling near-perfect restoration of occluded faces.

show abstract

Section: Compare With Other Methodsmentioning

confidence: 99%

Section: Compare With Other Methodsmentioning

confidence: 99%

Section: Quantitative Comparison Of Face Privacy-protection Effectmentioning

confidence: 98%

Section: Quantitative Comparison Of Face Privacy-protection Effectmentioning

confidence: 99%

See 2 more Smart Citations

Generation of Face Privacy-Protected Images Based on the Diffusion Model

You,

Zhao,

Wang

et al. 2024

Entropy

View full text Add to dashboard Cite

show abstract

“…Recent advances in generative machine learning such as variational autoencoders (VAEs) and generative adversarial networks (GANs) are starting to be applied to the anonymization problem [18][19][20][21][22][23][24]. The main idea of the generative approach is to learn the salient characteristics of the data distribution and sample new (synthetic) individuals from the distribution.…”

Section: Introductionmentioning

confidence: 99%

Preservation of Anomalous Subgroups On Variational Autoencoder Transformed Data

Maina¹,

Bryant²,

Ogallo³

et al. 2020

ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

In this paper, we investigate the effect of machine learning based anonymization on anomalous subgroup preservation. In particular, we train a binary classifier to discover the most anomalous subgroup in a dataset by maximizing the bias between the groups predicted odds ratio from the model and observed odds ratio from the data. We then perform anonymization using a variational autoencoder (VAE) to synthesize an entirely new dataset that would ideally be drawn from the distribution of the original data. We repeat the anomalous subgroup discovery task on the new data and compare it to what was identified pre-anonymization. We evaluated our approach using publicly available datasets from the financial industry. Our evaluation confirmed that the approach was able to produce synthetic datasets that preserved a high level of subgroup differentiation as identified initially in the original dataset. Such a distinction was maintained while having distinctly different records between the synthetic and original dataset. Finally, we packed the above end-to-end process into what we call Utility Guaranteed Deep Privacy (UGDP) system. UGDP can be easily extended to onboard alternative generative approaches such as GANs to synthesize tabular data.

show abstract

PICLAnony: Controlled Anonymization of Face Attributes Based on Parametric Imitation Comparison Learning

Liu,

Ji,

Liang

2024

Preprint

View full text Add to dashboard Cite

With the ubiquitous use of mobile devices and the widespread sharing of personal photos and videos on social media, the use of anonymized images to avoid identity disclosure has become essential. Aiming at the problem of low-quality and uncontrollable anonymous faces generated by existing anonymization algorithms, we propose PICLAnony, a controllable anonymization algorithm for face attributes based on parametric imitation comparison learning. It transfers the four visual information corresponding to identity, expression, pose, and illumination of the source image to the generated anonymized face image through parametric imitation contrast learning. And it edits these attribute features that reflecting sensitive behavioral intentions in a context-controlled manner. In the parameter imitation learning stage, high-quality and pose-controllable anonymized faces are generated by imitating the semantic parameters of real images. In the parameter comparison learning stage, the semantic parameters of the edited anonymized image are compared and learned with those of the source image, which solves the problem of insufficient decoupling of expression and illumination attributes in the editing process. In addition, a background control module is designed to keep the background controllable during the editing process of anonymous face facial attributes. We show the subjective results of our algorithm on both CelebA\_HQ and FFHQ datasets, and the subjective and objective results demonstrate that PICLAnony outperforms the state-of-the-art methods in terms of image quality and editing of facial attributes of anonymized faces.

show abstract

DeepPrivacy: A Generative Adversarial Network for Face Anonymization

Cited by 3 publications

References 26 publications

Generation of Face Privacy-Protected Images Based on the Diffusion Model

Generation of Face Privacy-Protected Images Based on the Diffusion Model

Preservation of Anomalous Subgroups On Variational Autoencoder Transformed Data

PICLAnony: Controlled Anonymization of Face Attributes Based on Parametric Imitation Comparison Learning

Contact Info

Product

Resources

About