This post is also available in:
עברית (Hebrew)
Protecting digital images from manipulation has become increasingly difficult. As AI-generated content and sophisticated editing tools become more accessible, verifying whether a photo is genuine is turning into a major challenge for journalists, forensic investigators, government agencies and security organizations. Many existing security systems rely on pseudo-random number generators (PRNGs), which create number sequences using mathematical algorithms. While suitable for many applications, these sequences are deterministic, meaning that under certain conditions they can be predicted or reproduced.
Researchers at Hanyang University have developed an alternative that generates randomness directly from a physical process rather than a mathematical formula. Their photospike-based true random number generator (PS-TRNG) uses pulses of light to produce unpredictable electrical signals, creating random numbers that can be embedded into digital images as a built-in authenticity marker.
According to TechXplore, the device consists of copper vanadate nanostructures and a layer of tin dioxide quantum dots sandwiched between a semiconductor substrate and a transparent conducting electrode. When the system is illuminated with pulsed red light, electrons travel through the structure toward the upper electrode. Along the way, some of these electrons become trapped at naturally occurring defect sites inside the quantum-dot layer. Because the locations and timing of this trapping process are inherently random, the electrical current generated by each light pulse is slightly different.
When the light is switched off, the trapped electrons are released, producing another unpredictable electrical signal. The system converts these continuously varying currents into three digital states—0, 1 and 2—creating ternary random numbers in real time. According to the researchers, the generator passed all 15 statistical tests in the National Institute of Standards and Technology (NIST) randomness test suite and maintained stable performance over more than two million operating cycles and 460 days of testing.
The team also demonstrated a practical application by embedding the generated random numbers into digital images as invisible security signatures. Because the signature originates from a physical process inside the hardware, any modification to the image alters the embedded fingerprint. This enables pixel-level detection of image manipulation, including AI-generated alterations, without affecting the visible appearance of the photograph.
Although the research focuses on image authentication, the technology could also have important defense and homeland security applications. Trusted imagery is essential for intelligence analysis, surveillance, forensic investigations and the verification of evidence. Integrating a hardware-generated security fingerprint directly into cameras or imaging sensors could help ensure that images remain verifiable from the moment they are captured, reducing the risk of tampering before they are used for operational or legal purposes.
The researchers believe this approach could shift image security from software-based verification performed after an image is created to hardware-level protection built directly into future imaging devices.
The research was published here.
