Yoti's Age Estimation Software Improves Accuracy with New ACCS Evaluation

Published Date : 29/10/2024 

Yoti's facial age estimation (FAE) software has seen significant improvements in accuracy, as confirmed by the latest evaluation from the Age Check Certification Scheme (ACCS). This development could enhance trust in age verification technology. 

Yoti, a leading provider of age assurance and ID verification solutions, has announced a major breakthrough in the accuracy of its facial age estimation (FAE) software. According to a report from the Age Check Certification Scheme (ACCS), Yoti's September 2024 model has shown substantial improvements in Mean Absolute Error (MAE) for 18-year-olds. The latest evaluation reveals an MAE of just 1.05 years, with a Standard Deviation (SD) of 1.01 years, marking a significant improvement from the 1.79 MAE reported in the November 2020 evaluation of Yoti's 2020 model.Matt Prendergast, Yoti's product marketing director, noted, “This reduction in MAE underscores our ongoing commitment to enhancing the performance of our age estimation technology. The latest ACCS results are a testament to our efforts to refine and improve our algorithms.”However, the ACCS evaluation results are notably different from those of the National Institute of Standards and Technology (NIST). NIST's recent evaluation of Yoti's September 2024 model showed an MAE of 2.63 years for visa images captured in Mexico consular offices and 2.90 years for images captured in US immigration offices. Prendergast emphasized that the source of test images significantly impacts FAE accuracy. “It's crucial to understand that NIST test images are not captured on mobile phones, which is the primary method used by end users for digital age checks. The latest ACCS testing demonstrates that this makes a substantial difference in accuracy,” he explained.Since 2018, Yoti has focused on training its FAE algorithms using mobile phone images, as the majority of users prefer this method. The company has also conducted its own internal evaluations, which found an MAE for 18-year-olds of 1.24 years. Yoti hopes that the lower MAE from the ACCS evaluation will help build trust among regulators who are still skeptical about age estimation technology.The ACCS results also indicate a slight decline in accuracy as the age buffer increases. While 99.3% of 18-year-olds are reliably estimated to be under 25 (a 7-year buffer), the reliability drops to 94.2% for a 2-year buffer, meaning 18-year-olds are reliably estimated to be under 20. Despite this, Prendergast highlighted, “These results show that a very high percentage of adults can use Yoti FAE to prove they are 18 or over. We have addressed concerns about accuracy and spoofing, and our published results by gender and skin tone demonstrate no significant bias.”Addressing the issue of bias in biometric technology is crucial for widespread adoption. Prendergast stated, “Claims of bias are often made against biometric tech, but standardized evaluations increasingly show that such concerns are unfounded.”Yoti CEO Robin Tombs has also weighed in on the ongoing debate surrounding age verification technology, particularly in the context of the US Supreme Court case between the Free Speech Coalition and Texas Attorney General Paxton. Paxton is pushing for age verification as a requirement for accessing adult content, as part of the controversial HB 1181 bill. Tombs criticized some of the misconceptions among bill opponents and supporters, particularly the claim that it’s easy to circumvent facial age estimation by replaying a video or displaying a photo to the camera. He rated this claim as a “9 on the whoppers scale.”Tombs emphasized, “The Texas law requires age assurance to be effective, and without liveness detection technology, the entire process would be pointless. Liveness detection is crucial to ensuring that the image presented to the camera is a real user, not a spoof.” He cited iBeta Quality Assurance's evaluations, which require vendors to detect 100% of screen and paper attacks to meet Level 1 standards and 99% of more challenging mask attacks (costing up to $300) to meet Level 2 standards.Tombs suggested that NIST should be called as a witness in the court case and proposed a challenge for critics to beat leading vendors' liveness detection technology. “This type of STEM experiment would be educational for both children and adults,” he said.As the debate over HB 1181 continues, Yoti remains committed to advancing age verification technology and building trust among regulators and the public. 

Frequently Asked Questions (FAQS):

Q: What is the Age Check Certification Scheme (ACCS)?

A: The Age Check Certification Scheme (ACCS) is a UK-based certification program that evaluates and certifies age verification systems. It ensures that these systems are accurate and reliable, providing a standardized way to measure the performance of age estimation technologies.

Q: What is Mean Absolute Error (MAE) in the context of age estimation?

A: Mean Absolute Error (MAE) is a metric used to measure the average magnitude of errors in a set of predictions, without considering their direction. In age estimation, it indicates the average difference between the estimated age and the actual age of individuals.

Q: Why are mobile phone images important for age estimation accuracy?

A: Mobile phone images are important because they are the most commonly used method for end users to capture their selfies for digital age checks. Training age estimation algorithms on mobile phone images helps improve the accuracy of the technology in real-world scenarios.

Q: What is liveness detection in biometric technology?

A: Liveness detection is a feature in biometric systems that ensures the biometric sample (e.g., a selfie) is from a real, live person and not a spoof or fake. It is crucial for preventing fraud and ensuring the security of age verification and identity systems.

Q: How does Yoti address concerns about bias in its age estimation technology?

A: Yoti addresses concerns about bias by publishing results that show no significant differences in accuracy across different genders and skin tones. The company conducts rigorous internal and external evaluations to ensure its technology is fair and unbiased. 

More Related Topics :