CNNs Under Attack: On the Vulnerability of Deep Neural Networks Based Face Recognition to Image Morphing

Wandzik, Lukasz; Garcia, Raul Vicente; Kaeding, Gerald; Chen, Xi

doi:10.1007/978-3-319-64185-0_10

Cited by 22 publications

(9 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After that, some other metrics including attack presentation classification error rate (APCER) and the bona fide presentation classification error rate (BPCER) are complemented to evaluate the biometric systems' vulnerability to morphing attacks [6]. In addition, the vulnerability of deep learning based FRS to morphing attack was investigated in [7], and it is found that morphing attacks can degrade the performance of FRS to a certain extent.…”

Section: A Vulnerability Of Frs To Face Morphing Attackmentioning

confidence: 99%

See 1 more Smart Citation

FD-GAN: Face De-Morphing Generative Adversarial Network for Restoring Accomplice’s Facial Image

2019

View full text Add to dashboard Cite

Face morphing attack is proved to be a serious threat to the existing face recognition systems. Although a few face morphing detection methods have been put forward, the face morphing accomplice's facial restoration remains a challenging problem. In this paper, a face de-morphing generative adversarial network (FD-GAN) is proposed to restore the accomplice's facial image. It utilizes the symmetric dual network architecture and two levels of restoration losses to separate the identity feature of the morphing accomplice. By exploiting the captured facial image (containing the criminal's identity) from the face recognition system and the morphed image stored in the e-passport system (containing both criminal and accomplice's identities), the FD-GAN can effectively restore the accomplice's facial image. The experimental results and analysis demonstrate the effectiveness of the proposed scheme. It has great potential to be applied for tracing the identity of face morphing attack's accomplice in criminal investigation and judicial forensics.INDEX TERMS Face de-morphing, face morphing attack, facial restoration, generative adversarial network.

show abstract

Section: A Vulnerability Of Frs To Face Morphing Attackmentioning

confidence: 99%

“…After this idea was put forward, some concerns about the vulnerability of commercial FRS with respect to morphed face attack have been investigated [3][4][5][6][7]. It was proved that face morphing attack is posing a serious threat to the existing FRS.…”

Section: Introductionmentioning

confidence: 99%

FD-GAN: Face De-Morphing Generative Adversarial Network for Restoring Accomplice’s Facial Image

2019

View full text Add to dashboard Cite

show abstract

“…Nonetheless, in their experiment, they use Psychomorph, which creates lower quality morphings (with more ghost artifacts) than GIMP+GAP. More examples of verification attacks can be found in L. Wandzik et al [42] and U. Scherhag [43]. In the first one, they carried out the experiment using FaceNet, utilizing more than 3000 pairs with 22 morphed images between each pair, working with triplets of images (impostor-accomplice-morphing).…”

Section: Spoofing Attacksmentioning

confidence: 99%

Stegano-Morphing: Concealing Attacks on Face Identification Algorithms

Carabe

Cermeño²

2021

IEEE Access

View full text Add to dashboard Cite

Face identification is becoming a well-accepted technology for access control applications, both in the real or virtual world. Systems based on this technology must deal with the persistent challenges of classification algorithms and the impersonation attacks performed by people who do not want to be identified. Morphing is often selected to conduct such attacks since it allows the modification of the features of an original subject's image to make it appear as someone else. Publications focus on impersonating this other person, usually someone who is allowed to get into a restricted place, building, or software app. However, there is no list of authorized people in many other applications, just a blacklist of people no longer allowed to enter, log in, or register. In such cases, the morphing target person is not relevant, and the main objective is to minimize the probability of being detected. In this paper, we present a comparison of the identification rate and behavior of six recognizers (Eigenfaces, Fisherfaces, LBPH, SIFT, FaceNet, and ArcFace) against traditional morphing attacks, in which only two subjects are used to create the altered image: the original subject and the target. We also present a new morphing method that works as an iterative process of gradual traditional morphing, combining the original subject with all the subjects' images in a database. This method multiplies by four the chances of a successful and complete impersonation attack (from 4% to 16%), by deceiving both face identification and morphing detection algorithms simultaneously.

show abstract

“…The proposed approach shows Bonafide Presentation Classification Error Rate (BPCER) value of 14.38%, 41.78%, and 28.76% at 5% Attack Presentation Classification Error Rate (APCER). Wandzik et al (2017) shows the vulnerability of deep CNN based face recognition under morphing attacks. ResNet v1 shows 99.97% acceptance rate on original images; the acceptance rate drops down to 34.66% on morphed images blended with 0.5% probability of images.…”

Section: Related Workmentioning

confidence: 99%

MagNet: Detecting Digital Presentation Attacks on Face Recognition

Singh

Vatsa

Noore

2021

Front. Artif. Intell.

View full text Add to dashboard Cite

Presentation attacks on face recognition systems are classified into two categories: physical and digital. While much research has focused on physical attacks such as photo, replay, and mask attacks, digital attacks such as morphing have received limited attention. With the advancements in deep learning and computer vision algorithms, several easy-to-use applications are available where with few taps/clicks, an image can be easily and seamlessly altered. Moreover, generation of synthetic images or modifying images/videos (e.g. creating deepfakes) is relatively easy and highly effective due to the tremendous improvement in generative machine learning models. Many of these techniques can be used to attack the face recognition systems. To address this potential security risk, in this research, we present a novel algorithm for digital presentation attack detection, termed as MagNet, using a “Weighted Local Magnitude Pattern” (WLMP) feature descriptor. We also present a database, termed as IDAgender, which consists of three different subsets of swapping/morphing and neural face transformation. In contrast to existing research, which utilizes sophisticated machine learning networks for attack generation, the databases in this research are prepared using social media platforms that are readily available to everyone with and without any malicious intent. Experiments on the proposed database, FaceForensic database, GAN generated images, and real-world images/videos show the stimulating performance of the proposed algorithm. Through the extensive experiments, it is observed that the proposed algorithm not only yields lower error rates, but also provides computational efficiency.

show abstract

CNNs Under Attack: On the Vulnerability of Deep Neural Networks Based Face Recognition to Image Morphing

Cited by 22 publications

References 19 publications

FD-GAN: Face De-Morphing Generative Adversarial Network for Restoring Accomplice’s Facial Image

FD-GAN: Face De-Morphing Generative Adversarial Network for Restoring Accomplice’s Facial Image

Stegano-Morphing: Concealing Attacks on Face Identification Algorithms

MagNet: Detecting Digital Presentation Attacks on Face Recognition

Contact Info

Product

Resources

About