For centuries, the lens has been the eye of every imaging device — a carefully shaped piece of glass that bends light into meaning. Now, researchers at Nanjing University and Peking University have proposed a different kind of seeing: one where the lens disappears entirely, replaced by computation, physics, and a neural network trained to reconstruct motion from a single captured moment. In doing so, they have quietly expanded what a camera can be, and where it can go.
Researchers develop lensless imaging breakthrough for sharp dynamic object capture
The lens, for all its elegance, has always been a constraint.
Mark
Why does motion make lensless imaging so much harder than regular photography?
Mimi
Because the sensor captures everything at once—all the diffraction patterns from every moment of motion layered on top of each other. A lens freezes a moment. A lensless sensor without a way to separate those moments just sees a blur.
Mark
So the mask helps by making the patterns more readable?
Mimi
Exactly. The binary mask acts like a translator. It modulates the light in a way that makes the diffraction patterns carry more information about what's actually there, rather than just being noise.
Mark
And the neural network learns to decode that?
Mimi
It learns the relationship between the patterns and the actual object—across space and time simultaneously. It's not just pattern-matching; it's learning the physics of how light moves.
Mark
Why does working in both spatial and frequency domains matter?
Mimi
The spatial domain tells you where edges are. The frequency domain tells you about fine details and texture. Motion blur lives in the frequency domain. By constraining the reconstruction in both, you suppress the blur while keeping the sharpness.
Mark
Can this actually work on living things, or just test targets?
Mimi
They tested it on swimming rotifers—real organisms moving in unpredictable ways. The system captured tail contractions and body displacement. That's not a controlled scenario. That's biology happening.
Mark
What does this mean for someone carrying a medical device?
Mimi
It means the device gets smaller, lighter, and doesn't sacrifice image quality. You're not trading clarity for portability anymore.
O Pulso
- Lensless imaging promised smaller, lighter devices for medicine and microscopy, but motion had always broken the system — a moving object turned sensor data into an unsolvable blur.
- Every existing workaround demanded multiple image captures or rigid assumptions about the world, making real-time biological observation effectively impossible.
- The McLDI-INR framework attacks the problem from two directions at once — a physical mask that enriches incoming light patterns, and a neural network that learns to reconstruct moving objects across both space and time from a single shot.
- Tested on real microscopic animals — rotifers twisting and contracting freely — the system recovered sharp edges and fine structural details that conventional lensless methods could not.
- The trajectory points toward imaging devices that no longer need a lab to function: pocket microscopes, wearable diagnostics, and biological sensors freed from the tyranny of glass.
For centuries, the lens has been the eye of every imaging device — a carefully shaped piece of glass that bends light into meaning. Now, researchers at Nanjing University and Peking University have proposed a different kind of seeing: one where the lens disappears entirely, replaced by computation, physics, and a neural network trained to reconstruct motion from a single captured moment. In doing so, they have quietly expanded what a camera can be, and where it can go.
The traditional optical lens, that centuries-old piece of carefully ground glass, is becoming an obstacle. As devices shrink — phones, wearables, portable medical scanners — the lens takes up space it can no longer justify. Lensless imaging offers an alternative: a sensor captures raw diffraction patterns from light bouncing off an object, and a computer reconstructs the image that a lens would have produced. Smaller, lighter, and theoretically sharper.
The problem has always been motion. When an object moves, overlapping diffraction patterns pile up into an unreadable blur. Existing methods either demanded multiple exposures or leaned on assumptions too rigid for the living world. For anyone hoping to watch cells divide or organisms swim, lensless imaging had hit a wall.
Researchers at Nanjing University and Peking University have now broken through it. Their system, published in June 2026, places a binary mask between the object and the sensor to enrich the captured light patterns, then deploys a deep neural network that learns to reconstruct a moving object's shape and position across continuous space and time — from a single measurement. Crucially, the reconstruction is constrained not just at the sensor plane but also in the frequency domain, using the actual physics of light to recover details that motion blur typically destroys.
The team validated their approach on simulated moving targets and then on real hardware: a resolution test chart sliding across the field of view, and freely swimming rotifers — microscopic animals whose bodies twist, contract, and displace in ways that defeat most computational imaging systems. In both cases, the method recovered sharp edges and fine structural detail where only blur had existed before.
What opens up beyond the technical result is the more consequential story. Miniaturized microscopes small enough for a smartphone. Portable biological detection that requires no laboratory. Wearable medical devices capable of genuine visual clarity. The lens was always elegant — but it was also always a constraint. There is now a credible path around it.
The camera as we know it is becoming a problem. As phones grow thinner, as wearable devices proliferate, as portable medical scanners need to fit into smaller spaces, the traditional lens—that carefully ground piece of glass that has defined imaging for centuries—starts to feel like dead weight. It bends light, yes, but it also bends the possibilities of what a device can be. It introduces aberrations. It scatters colors. It takes up room.
Lensless imaging offers a different path. Instead of a lens, you have a sensor that captures the raw diffraction patterns of light bouncing off an object. The image doesn't exist yet—it's encoded in those patterns, a mathematical puzzle waiting to be solved. A computer reconstructs what the lens would have shown, but without the lens itself. The result is smaller, lighter, and theoretically sharper.
But there's a catch. When the object moves, everything falls apart. The sensor records a blur of overlapping diffraction patterns, all mixed together. Conventional methods need multiple shots to untangle the mess, or they rely on assumptions so strict they miss the real world. For microscopy, for watching living cells, for any situation where something is actually happening—lensless imaging has been stuck.
Researchers at Nanjing University and Peking University have broken through that wall. Their method, published in June 2026 in the journal Intelligent Opto-Electronics, combines two ideas: a physical model of how light actually behaves, and a deep neural network trained to learn from that behavior. They call it McLDI-INR, which stands for Mask-constraint Lensless Dynamic Imaging by Dual-Domain Collaborative Implicit Neural Representation. The name is dense, but the idea is elegant.
The system works like this. A binary mask sits between the object and the sensor, modulating the incoming light into patterns that carry more useful information. The neural network then takes spatial and temporal coordinates—where something is, and when—and learns to reconstruct the complex amplitude of the moving object across continuous space and time. Crucially, the framework doesn't just constrain the reconstruction at the sensor plane. It also works in the frequency domain, using physics-based rules to recover the fine details that motion blur typically erases.
In simulations, the method proved itself on rapidly moving targets, both linear and nonlinear. It preserved object edges, contours, and texture details far better than existing approaches. But simulation is one thing. The researchers built an actual lensless imaging system and tested it on real objects: a USAF resolution target moving across the field, and freely swimming rotifers—microscopic animals that twist and contract as they move. The results showed sharp edges where there had been blur before. In the biological samples, the system captured the fine details of the rotifers' motion—tail contractions, body displacement—the kind of non-rigid movement that usually defeats computational imaging.
What matters here is not just the technical achievement, though that is real. It's the opening it creates. Miniaturized microscopes that fit into a phone. Portable biological detection systems that don't require a lab. Wearable medical devices that can actually see what they're looking at. The lens, for all its elegance, has always been a constraint. Now there's a way forward without it.
Citações Notáveis
The deep integration of physical modeling and implicit neural representation can expand the capability boundaries of lensless imaging, enabling high-fidelity reconstruction of dynamic scenes without relying on conventional optical lenses.— Research team, Intelligent Opto-Electronics
Quer a matéria completa?
Leia o original em News-Medical ↗
Health & Medicine