NIST, the National Institute of Standards and Technology, recently published an article that demonstrated how face morphing software is used to commit identity fraud. This software combines two photos of different people into a singular image. If given to the wrong hands, a bad actor could potentially bypass identity verification systems in secure locations such as airports or at country borders. These photos have the ability to deceive face recognition systems, misidentifying the image as both individuals, allowing an individual to pose as both the first and second individual.
Face Analysis Technology Evaluation Morph
NIST released Face Analysis Technology Evaluation (FATE) Morph to help combat this ongoing issue. By building awareness of this issue, NIST can effectively guide organizations to detect morphing where photograph identification is a concern. The article contains guidelines for possible remediation strategies if a morphed photograph is detected. While the single image detection model posts an accuracy of 100% with a false detection rate of 1% on single image detection, detection accuracy can drop to 40% with morphed photos generated by software not recognized by FATE. Differential detectors provide an accuracy ranging between 72% and 90% across morphs that are created with open-source and closed-source morphing software. The difference between the detectors is explained below.
S-MAD (Single Image Morph Attack Detectors)
S-MAD operates on a suspected morphed photograph. This detector applies image processing and pattern classification techniques to detect blemishes and artifacts that are created as a result of morphing images.
D-MAD (Differential Morph Attack Detectors)
D-MAD operates on a suspected morph and a genuine photo for comparison. These detectors are more consistent because they are built using technology that is primarily used for face recognition.
The Potential Future of Morph Detection
As technology becomes more advanced, bad actors could leverage artificial intelligence to develop more realistic face morphs that can continue to deceive detection tools. Although the detection software is continuously improving, so is the sophistication of open and closed source face morphing software. Bad actors could potentially target the detection vulnerabilities of the software. For example, a bad actor could add or remove certain artifacts from a morphed image to bypass software detection standards. It is important to note that the best procedure for investigating a potential morph includes a combination of human review, the use of automated tools, and a specific process to review suspected morphs.
How to Remain Safe From Morphs
- Require candidates and employees to take identification pictures on-site to prevent any chance for alteration
- Deploying automated morph detection in secure areas can significantly reduce the risk of unauthorized personnel accessing a secure area
- Train and educate users to detect certain artifacts or blemishes within a suspected morphed image
Sources:
https://pages.nist.gov/frvt/reports/morph/fate_morph_4B_NISTIR_8584.pdf
