UCLA Built a Mirror That Lies. Point Anything at It and the Reflection Comes Back as the Number 8.
Aydogan Ozcan's group used a structured diffractive surface — not software — to overwrite whatever an unknown object looks like with one predetermined image.
Engineers at UCLA have built a mirror that reflects something other than what is in front of it. Hold up an object — any object, including one the device has never encountered — and the light coming back carries a single predetermined image instead. In the group's demonstration, everything came back as the numeral 8.
The work, published in Nature Communications, comes out of the UCLA Engineering Institute for Technology Advancement, where the research was led by Aydogan Ozcan, with Yuhang Li as lead author. The device pairs an ordinary reflective mirror with a structured diffractive surface — a thin, precisely patterned optic that shapes the phase of light passing through it. Nothing about the transformation happens in software after the fact.
"Instead of simply distorting a reflection, the lying mirror is designed to optically replace the visual information carried by many different, unknown inputs with a predefined deceptive pattern," Ozcan said.
The distinction matters. A digital system that alters an image has to capture it first, which means the true image exists somewhere, if only for a moment, and can in principle be intercepted. The lying mirror never forms it. The scrambling is performed by passive light-matter interaction at the surface itself, at the speed of light, with no power draw and no computation at the moment of use. What can be read out downstream is only the decoy.
Getting a surface to do that required training. The team used deep learning to optimize the diffractive layer's phase profile, searching for a pattern that maps an enormous and open-ended space of possible inputs onto one fixed output. The result was validated on an optical bench using a programmable micro-mirror array to generate test objects, illuminated at 600, 550 and 480 nanometers — red, green and blue. Crucially, the transformation held up when inputs were randomly rotated, shifted, rescaled or contaminated with noise, which is the difference between a laboratory curiosity and something that could survive contact with an uncontrolled scene. A separate broadband design extends the effect across continuous spectral ranges rather than three discrete lines.
The obvious applications run toward security, defense and anti-surveillance: a surface that defeats optical inspection without announcing that it is doing so, since what an observer sees is a coherent image rather than an obviously corrupted one. Ozcan's group also points to entertainment and display uses. The broader significance is architectural. Diffractive optical processors of this kind — passive, all-optical, trained offline and then frozen into a physical surface — are being pushed as a way to move classes of computation out of electronics entirely. This one performs an information-destroying operation, which is a harder thing to do well than it sounds: the design has to guarantee that many different inputs collapse to the same output, and that no residue of the original survives in the reflected field.
Originally reported by Phys.org.