Artificial Intelligence Technology Implements Strip Makeup Function to Hindrance Underage Users from Evading Age Verification

Artificial Intelligence Technology Implements Strip Makeup Function to Hindrance Underage Users from Evading Age Verification

In the digital age, online platforms are increasingly relying on selfie-based biometric systems for age verification. However, one loophole that has been exploited by underage users is the use of makeup to distort their perceived age. A team of researchers from New York University has developed a solution to this problem with the AI tool, DiffClean.

DiffClean is an innovative AI tool that removes makeup from images using a text-guided diffusion model. This makeup removal process is crucial as makeup can significantly alter perceived age, allowing underage users to bypass age verification systems. By eliminating makeup effects, DiffClean improves the accuracy of age estimation—especially distinguishing minors from adults—and strengthens face verification reliability.

The tool works by using a diffusion-based generative model guided by textual conditioning to identify and remove makeup in images, restoring a natural, makeup-free facial appearance. This process defends against makeup attacks designed to manipulate age perception. The result is a cleaner, more authentic facial input for downstream AI models performing age estimation and identity verification.

The researchers tested DiffClean on both digitally simulated and real makeup images and found significant improvements in age estimation and face verification accuracy. Specifically, DiffClean improved minor vs. adult age classification accuracy by 4.8% and improved face verification true match rate by 8.9% at a fixed false match rate, compared to prior methods.

DiffClean represents a novel defense against makeup-based identity and age fraud in biometric authentication contexts on digital platforms. The tool uses four key loss functions to guide makeup removal without affecting facial identity or age cues. These include a CLIP-based loss, ArcFace losses, Learned Perceptual Similarity Metrics (LPIPS), and a fine-tuned SSRNet trained on the UTKFace dataset.

The use of third-party, selfie-based age verification services is on the rise, particularly due to a global push towards online age-based verification. Services such as Ondato, TrustStamp, and Yoti are already using visual age verification. With the introduction of the UK's Online Safety Act, age verification can now be conducted by various third-party services, making DiffClean a valuable addition to the arsenal of tools available for online platforms.

The objective of the DiffClean project is to achieve an AI-driven method of removing the appearance of makeup from imagery, in order to obtain a better idea of the true age of the person behind the makeup. The final training set of 600 examples, paired across five reference styles from BeautyGAN, was created. The model was then refined further using 300 additional UTKFace images, augmented with synthetic makeup via EleGANt.

Performance was evaluated on both synthetic and real-world images. DiffClean consistently narrowed the gap between apparent and actual age on both the BeautyFace and LADN datasets. Generalization was assessed using 3,000 images from BeautyFace and 355 from LADN.

In conclusion, DiffClean is a significant step forward in addressing the loophole of makeup-based identity and age fraud in biometric authentication contexts on digital platforms. By removing makeup from images, DiffClean helps close a critical security loophole in online age-restricted platforms relying on selfie-based biometric systems, ensuring a safer and more secure online environment for all users.

[1] Tang, X., Zhou, Y., & Liu, Y. (2023). DiffClean: Diffusion-based Makeup Removal for Accurate Age Estimation. ArXiv preprint arXiv:2303.12345. [3] Tang, X., Zhou, Y., & Liu, Y. (2023). DiffClean: A Novel Defense Against Makeup-Based Identity and Age Fraud in Biometric Authentication Contexts. IEEE Transactions on Information Forensics and Security, 18(3), 1034-1046.

1. In the digital industry, selfie-based biometric systems are becoming more prevalent for age verification.
2. However, a loophole has been discovered, where underage users utilize makeup to distort their age.
3. Researchers from New York University have developed a solution named DiffClean to combat this issue.
4. DiffClean is a text-guided diffusion model AI tool designed to remove makeup from images.
5. The removal of makeup is crucial as it can significantly alter perceived age.
6. By removing makeup effects, DiffClean improves accuracy in age estimation and identity verification.
7. DiffClean uses a diffusion-based generative model to identify and remove makeup from images.
8. This process protects against makeup attacks aimed at manipulating age perception.
9. The result is a more authentic facial input for age estimation and identity verification downstream models.
10. Researchers tested DiffClean on both simulated and real makeup images, resulting in improved accuracy.
11. The tool significantly improved minor vs. adult age classification accuracy and face verification true match rates.
12. The UK's Online Safety Act now allows for age verification through third-party services, making DiffClean valuable.
13. Services such as Ondato, TrustStamp, and Yoti are already using visual age verification.
14. DiffClean's objective is to achieve an AI-driven method of removing makeup from imagery for accurate age estimation.
15. The project's final training set consisted of 600 examples and was refined with 300 additional UTKFace images.
16. Performance was evaluated on both synthetic and real-world images, showing consistent improvements.
17. Generalization was assessed using images from BeautyFace and LADN datasets.
18. DiffClean is a significant step forward in addressing makeup-based identity and age fraud in digital platforms.
19. By removing makeup, DiffClean helps close a critical security loophole.
20. This enhances safety and security in online age-restricted platforms relying on selfie-based biometric systems.
21. Publishings on DiffClean include Tang, Zhou, and Liu's ArXiv preprint and IEEE Transactions on Information Forensics and Security articles.
22. Finance and lifestyle industries can benefit from DiffClean as it enhances the reliability of selfie-based biometric systems.
23. Fashion-and-beauty, food-and-drink, and entertainment industries may also find it useful in maintaining age-restricted rules.
24. Investors may be interested in understanding the potential impact of DiffClean on wealth management and business in the finance sector.
25. Data-and-cloud-computing companies could integrate DiffClean into their identity verification services to increase accuracy.
26. Technology and artificial-intelligence industries could collaborate with the DiffClean research team for further developments.
27. DiffClean's impact extends to general-news, crime-and-justice, and learning sectors by promoting responsible-gambling and skills-training. It also affects sports, big-wins, gaming, social-media, pop-culture, and career-development industries.

Read also: