Generating Master Faces for Dictionary Attacks with a Network-Assisted Latent Space Evolution

Shmelkin, Ron; Friedlander, Tomer; Wolf, Lior

doi:10.1109/fg52635.2021.9666968

Cited by 20 publications

(35 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Master Face Attacks on Face Recognition Systems [11] & Master Faces for Dictionary Attacks [15]. Recent work has suggested that it is possible to bypass facial verification systems with a single master face.…”

Section: Related Workmentioning

confidence: 99%

“…Recent work has suggested that it is possible to bypass facial verification systems with a single master face. The authors of [15] used a greedy coverage search to find an image within the specific dataset that consisted of the most similar features. However, the master faces have been generated using biased training data (LFW) and hence are not very accurate (only successful for 40% of test data) and are specifically male Caucasians with white hair.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Face Verification Bypass

Sarda¹

2022

Preprint

View full text Add to dashboard Cite

Face verification systems aim to validate the claimed identity using feature vectors and distance metrics. However, no attempt has been made to bypass such a system using generated images that are constrained by the same feature vectors. In this work, we train StarGAN v2 to generate diverse images based on a human user that have similar feature vectors yet qualitatively look different. We then demonstrate a proof of concept on a custom face verification system and verify our claims by demonstrating the same proof of concept in a black box setting on dating applications that utilize similar face verification systems.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Face Verification Bypass

Sarda¹

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…In face recognition (FR), a MasterFace is a face image that can be successfully matched against a large portion of the population [23]. Such a face can be used to imperson-Figure 1: Visualization of theoretical concepts -Assuming an embedding space (on a unit-sphere) with perfect identityseparation, the three concepts are visualized for two and three-dimensional embeddings for a fixed decision threshold r = 0.4.…”

Section: Introductionmentioning

confidence: 99%

“…Consequently, MasterFaces attracted attention in both media and academia. For instance, the work from Shmelkin et al [23] won the Google Best Paper Award at IEEE International Conference on Automatic Face and Gesture Recognition 2021 for their work on generating MasterFaces. In that work, the authors report that a single face can cover more than 20% of the identities in the used database (for a given FR model at a threshold for a false match rate of 10 −3 ).…”

Section: Introductionmentioning

confidence: 99%

“…Contrarily to previous works, our experiments (a) make use of six stateof-the-art FR models, (b) are based on cross-dataset and cross-model evaluations, and (c) utilize testing datasets that are of a significantly higher size and variance. To ensure comparability with previous works [23,17], we use the original MasterFace images reported in the Arxiv versions of these papers (CC-BY-4.0) for the experiments. The empirical investigations showed that only an insignificant proportion of the testing identities are successfully matched against the MasterFaces and that the comparison scores resulting from comparisons with MasterFaces closely resembles the score distribution of zero-effort imposter.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

On the (Limited) Generalization of MasterFace Attacks and Its Relation to the Capacity of Face Representations

Terhörst¹,

Bierbaum²,

Huber³

et al. 2022

Preprint

View full text Add to dashboard Cite

A MasterFace is a face image that can successfully match against a large portion of the population. Since their generation does not require access to the information of the enrolled subjects, MasterFace attacks represent a potential security risk for widely-used face recognition systems. Previous works proposed methods for generating such images and demonstrated that these attacks can strongly compromise face recognition. However, previous works followed evaluation settings consisting of older recognition models, limited cross-dataset and cross-model evaluations, and the use of low-scale testing data. This makes it hard to state the generalizability of these attacks. In this work, we comprehensively analyse the generalizability of MasterFace attacks in empirical and theoretical investigations. The empirical investigations include the use of six state-of-the-art FR models, cross-dataset and cross-model evaluation protocols, and utilizing testing datasets of significantly higher size and variance. The results indicate a low generalizability when MasterFaces are training on a different face recognition model than the one used for testing. In these cases, the attack performance is similar to zero-effort imposter attacks. In the theoretical investigations, we define and estimate the face capacity and the maximum MasterFace coverage under the assumption that identities in the face space are well separated. The current trend of increasing the fairness and generalizability in face recognition indicates that the vulnerability of future systems might further decrease. Future works might analyse the utility of MasterFaces for understanding and enhancing the robustness of face recognition models.

show abstract