A new imaging technology developed by researchers at the University of Rochester could make it possible to see more clearly through deep tissue, dense fog, and other environments that scatter light. The system combines near-infrared imaging, ultrafast optical “shuttering,” and inexpensive silicon detectors to address two persistent problems in modern imaging: the loss of image quality caused by scattering and the high cost of specialized infrared sensors. In a second study, the researchers joined the optical method with artificial intelligence to expand the system’s field of view, potentially broadening its usefulness in medical imaging, autonomous vehicles, and other technologies that must operate when ordinary cameras struggle.
Near-infrared light is already central to a wide range of applications. Its longer wavelengths generally scatter less than visible light, allowing it to penetrate biological tissue more effectively and travel farther through certain obscuring environments. Medical researchers use near-infrared signals to investigate structures beneath the skin, while LiDAR systems use them to measure distance and construct three-dimensional maps for autonomous vehicles. Yet near-infrared light is not immune to scattering. When photons encounter tissue, fog, dust, or other complex materials, they can be deflected from their original paths. The resulting image may contain a mixture of direct and scattered photons, producing haze, blur, and misleading visual information.
The Rochester team’s approach is designed to reject much of that unwanted scattered light before it reaches the detector. The key is a technique known as time-gating, which functions like a camera shutter operating at extraordinary speed. Instead of mechanically opening and closing, the system uses precisely timed bursts of light to control when near-infrared information is allowed to pass through the imaging pathway. The active window lasts for approximately one picosecond, or one trillionth of a second. At that timescale, light travels only a fraction of a millimeter, allowing the system to distinguish photons arriving from different optical paths and suppress signals that have been delayed or redirected by scattering.
“In a traditional camera, the shutter is mechanical—when it opens, light comes in, and when it closes, light is rejected,” says Yang Xu, a University of Rochester doctoral researcher and lead author of the work. “In this case, we use light to control light.” The distinction is crucial. A mechanical shutter cannot operate fast enough to separate photons that have taken slightly different routes through a scattering medium. An optical gate, however, can be synchronized with ultrafast pulses and switched on and off on a timescale relevant to photon transport. This gives the imaging system a way to isolate the earliest, most direct components of a signal while reducing the contribution from diffuse light.
At the heart of the system is a thin film of indium tin oxide, a conductive and optically active material commonly used in technologies such as touchscreens. In the Rochester design, the film participates in the rapid conversion of near-infrared photons into visible light. This process, known as optical upconversion, shifts information from a spectral region that normally requires specialized detectors into one that can be recorded by conventional silicon-based sensors. Silicon detectors are widely manufactured, relatively inexpensive, and highly developed, but they are not naturally efficient at detecting many near-infrared wavelengths. By converting the incoming signal before detection, the researchers can use these accessible components without abandoning the advantages of near-infrared illumination.
The material’s unusual optical behavior is associated with an epsilon-near-zero regime, in which the effective dielectric response approaches zero at particular wavelengths. Under these conditions, light can interact with the material in ways that differ sharply from its behavior in ordinary optical media. The response can enhance nonlinear interactions and enable rapid modulation of transmitted or converted light. In the Rochester system, that behavior supports the ultrafast gate and the real-time conversion of near-infrared information into a visible image. The result is not simply a detector that sees a different color; it is an integrated strategy for controlling when information is accepted and how it is translated into a form that conventional imaging hardware can process.
The researchers report that the combination of time-gating and upconversion produces clearer images through difficult environments while avoiding the expense associated with many specialized near-infrared cameras. The potential applications are broad. In biomedical imaging, improved rejection of scattered photons could help researchers and clinicians distinguish structures beneath the surface of tissue, an important challenge in efforts to visualize tumors and other abnormalities. In autonomous navigation, a system that can maintain useful image information in fog or other scattering conditions could complement existing LiDAR and camera technologies. The approach may also be relevant to remote sensing, industrial inspection, and scientific experiments in which a target is hidden behind or within a visually complex material.
The first imaging advance did, however, involve a trade-off: the time-gated system could capture only a limited field of view. That limitation arises because the ultrafast optical arrangement and the associated detection geometry are optimized for a restricted region. A clear image of a small area can still be valuable, but many practical systems must monitor a larger scene. To address this challenge, the Rochester researchers collaborated with colleagues at UCLA and incorporated machine learning into the imaging pipeline. Their method uses computational reconstruction to infer a broader target area from the information captured by the optical system, effectively extending the visible scene beyond the region directly recorded at full scale.
The artificial-intelligence component does not replace the optical gate. Instead, it works with the measurements produced by the time-gated upconversion system. A trained computational model can learn how the restricted image relates to a larger scene and reconstruct information that would otherwise be lost because of vignetting, the darkening or reduction of illumination toward the edges of an image. The resulting approach, described as hybrid deep reconstruction, is intended to provide vignetting-free upconversion imaging through scattering materials. “Before applying artificial intelligence, we could see only a limited field of view,” Xu says. “By adding our collaborators’ methods, we can essentially reconstruct a much larger target area, enlarging the field of view our ultrafast time-gating technique can capture.”
Together, the two developments illustrate a broader shift in optical engineering: difficult imaging problems are increasingly being solved by combining materials science, ultrafast physics, detector technology, and computation rather than relying on a single component. The indium tin oxide film supplies a rapid optical conversion and gating mechanism; silicon detectors provide a lower-cost route to recording the signal; and machine learning helps compensate for geometric limitations in the hardware. The work was led by Robert Boyd, the William F. Krupke Distinguished Professor in Optics, whose laboratory has refined time-gating methods for more than a decade. University of Rochester collaborators included optics alumna Saumya Choudhary and physics doctoral student Long Nguyen. Supported by the U.S. National Science Foundation, the Office of Naval Research, and the U.S. Department of Energy, the research points toward imaging systems that could remain informative where conventional visible-light cameras—and even existing near-infrared instruments—are overwhelmed by scattering.
Subject of Research: Ultrafast near-infrared imaging, optical upconversion, time-gating, scattering suppression, and AI-assisted image reconstruction.
Article Title: Hybrid deep reconstruction for vignetting-free upconversion imaging through scattering in epsilon-near-zero materials
Web References: University of Rochester: https://www.rochester.edu/ ; Article DOI: https://doi.org/10.1038/s41377-026-02375-6 ; Related University of Rochester research announcement: https://doi.org/10.1038/s41467-026-71039-1
References: Light: Science & Applications, DOI: 10.1038/s41377-026-02375-6; related Nature Communications paper, DOI: 10.1038/s41467-026-71039-1
Image Credits: URochester photo / J. Adam Fenster
Keywords: Near-infrared imaging, optical upconversion, time-gated imaging, indium tin oxide, epsilon-near-zero materials, light scattering, deep tissue imaging, medical imaging, LiDAR, autonomous vehicles, silicon detectors, machine learning, artificial intelligence, nonlinear optics.
Tags: artificial intelligence in optical systemsautonomous vehicle sensingbiomedical imaging advancementscost-effective silicon detectorsdeep tissue imagingenhanced field of view in imagingfog and dust penetrationimaging through scattering environmentsinnovative imaging for medical and navigation applicationsnear-infrared light technologyoptical ultrafast shutteringovercoming light scattering in complex media
