As facial recognition gradually becomes more prevalent in digital applications, a research team comprising HEC Lausanne (Unil), HEIG-VD, and the Swiss startup Rayshaper SA has tackled these challenges by developing the FaceShield prototype. This solution aims to manage the use of facial images by artificial intelligence systems when they circulate on digital platforms, and to limit their use for identification or profiling purposes.
Posting a photo on a website, social media platform, or institutional web pages has become a common practice. Yet today, these images can be automatically analyzed by artificial intelligence systems capable of identifying a person, linking their face to other online content, and deriving a range of personal—and sometimes sensitive—information. A reality often invisible to those concerned, but very real from a technological standpoint: facial recognition systems now compare images to databases that may contain millions of faces.
It is this gray area—between everyday use and the automated exploitation of images—that FaceShield addresses. Developed with support from the [SEAL] (Security and Trust Innovation Program)—a Vaud-based innovation initiative led by Unil, EPFL, and HEIG-VD—this project has given rise to a solution that explores new ways to protect faces not from human eyes, but from machines.
“Individuals today have very few means to control how their images are used by automated systems. FaceShield explores a technical solution to this imbalance,” explains Yash Raj Shrestha, professor at HEC Lausanne. “This solution offers a concrete and proactive response to users by giving them back full control over their biometric data,” adds Valentina Ebrahimi, project manager at RayShaper SA and an alumna of HEC Lausanne.
Protecting faces without erasing them
Unlike traditional blurring or masking techniques, FaceShield relies on a more subtle approach. The system modifies facial features in a way that is quite imperceptible to the human eye, while making the image difficult for facial recognition algorithms to process. The photos remain readable, shareable, and understandable, but their value to machines is significantly reduced.
The original versions of the images can, if necessary, remain accessible only to authorized individuals. This approach allows for the separation of the public image—that which circulates online—from the source image, which is protected and controlled.
The envisaged use cases apply to both the private and professional spheres, such as the publication of expert portraits on institutional websites, corporate communications, and the media.
By exploring these questions, the project highlights a central challenge of digital transformation: how can we continue to produce and share visual content while limiting its invisible exploitation by automated systems?
Rethinking the role of images in digital environments
“By concretely examining how images are produced, shared, and analyzed by automated systems, the project helps shape new approaches to visual data protection,” concludes Professor Shrestha. This work is now informing scientific, regulatory, and societal discourse on recognition technologies and the responsible use of images.
HEC Lausanne contributed to the analysis of uses and societal challenges; RayShaper SA, based in Crans-Montana (VS) with offices in Pully (VD), designed the core algorithms, their integration and implementation; while HEIG-VD provided its expertise in cybersecurity and designed and implemented the conditional access modules.
More information on how the service works:
