A transparent object can be physically present yet nearly invisible to an ordinary camera. Glass, plastic films and living cells often produce little change in image brightness, even though they alter the phase of light passing through them. That hidden information can reveal an object’s shape, optical thickness and position—but recovering it becomes dramatically harder when the object is trapped between moving, light-scattering layers. Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have now demonstrated a way to reconstruct all of those properties from a single intensity measurement.
The technique, developed by a team led by Professor Mooseok Jang of KAIST’s Department of Bio and Brain Engineering, is designed for phase objects: transparent structures that modify the timing, or phase, of light without necessarily creating a visible brightness pattern. In phase imaging, these subtle changes are converted into information about morphology and optical thickness. If either the material’s refractive index or its physical thickness is known, the remaining quantity can also be estimated. This makes phase imaging valuable for inspecting transparent semiconductor and display components and for observing living cells without staining them.
The problem becomes particularly severe when scattering layers surround the object. A scattering layer can scramble the direction and phase of light, much like fog, frosted glass or a diffusive film. When the layers are dynamic and continuously move, the optical transmission changes from moment to moment. Light that carries information about the object may arrive at the detector as a complex, blurred pattern, while the scattering environment itself remains unknown. Conventional approaches often address this difficulty by acquiring multiple exposures, calibrating the scattering layers in advance or training artificial-intelligence systems on large collections of experimentally generated examples.
In the KAIST approach, the object is positioned between two dynamic scattering layers, identified as S1 and S2. The first layer lies in front of the object and the second behind it. Under broad, plane-wave illumination, the researchers found that the recorded intensity did not clearly reveal an object-related diffraction pattern. The scattering layers effectively erased the recognizable signature needed for reconstruction. The team therefore changed the illumination geometry, concentrating light onto a small spot on the first scattering layer rather than illuminating the layer uniformly.
This focused, point-like illumination is central to the method. By directing light through a localized region of the first dynamic layer, the researchers created conditions in which the light emerging from the system retained a more reliable connection to the transparent object. The resulting detector image remained partially blurred, but it contained a discernible diffraction pattern encoded by the object. That pattern provided the physical clues needed to infer how the object had modified the wavefront before the light encountered the second scattering layer.
The reconstruction system combines an optical model with an artificial-intelligence framework. The optical model describes how a light field changes as it propagates through the object and the two scattering layers, including the phase shift introduced by the object and the distortions caused by scattering. Instead of learning a direct mapping from thousands of labeled images, the AI system works within the constraints imposed by those physical laws. It searches for the object and scattering conditions most capable of producing the single measured intensity pattern.
This physics-guided strategy allows the system to estimate several unknowns at once. From one camera measurement, the researchers reconstructed the transparent object’s shape and optical thickness, characterized the blur associated with the scattering environment and determined the object’s position between the layers. In effect, the method reverses the journey of light: it begins with the scrambled intensity recorded by the detector and mathematically works backward to identify the wavefront and object that generated it.
The achievement could be significant for applications in which repeated measurements are impractical or the optical environment cannot be stabilized. Semiconductor and display manufacturing increasingly relies on transparent films, coatings and microstructures whose defects may be difficult to detect with brightness-based imaging. A single-shot system could potentially inspect such components without requiring the scattering environment to remain unchanged between exposures. In biomedicine, the same principle may eventually support label-free imaging of cells or tissues in optically complex surroundings.
Professor Jang described the work as the first demonstration of restoring both the shape and position of a transparent object from a single measurement while the object is fully enclosed by dynamic scattering layers. The team, whose study was co-first-authored by Yoosun Kim and Gookho Song, plans to extend the method to more complex scattering environments and practical imaging conditions. Published in Optica, the research suggests that even when light appears to have been destroyed by motion and diffusion, carefully designed illumination and physics-based computation can recover information that conventional cameras cannot see.
Subject of Research: Single-shot phase imaging and reconstruction of transparent objects hidden between dynamic scattering layers
Article Title: Single-shot imaging of phase objects fully enclosed by dynamic scattering layers
News Publication Date: August 6
Web References: https://doi.org/10.1364/OPTICA.593328
References: Kim, Y., Song, G., Jang, M. et al., “Single-shot imaging of phase objects fully enclosed by dynamic scattering layers,” Optica, published July 1, 2026. DOI: 10.1364/OPTICA.593328
Image Credits: KAIST
Keywords
Phase imaging, transparent objects, dynamic scattering, optical scattering, computational imaging, single-shot imaging, artificial intelligence, diffraction, semiconductor inspection, biomedical imaging, KAIST, Optica
