KAIST develops single-shot technique to image transparent objects hidden behind fog

Restoring objects between two moving scattering layers

Shape, thickness, position and scattering degree captured simultaneously

From left: Jang Mu-seok, a professor in KAIST's Department of Bio and Brain Engineering; Song Guk-ho, a doctoral candidate in the same department; and Kim Yu-seon, a master's candidate in the department. [KAIST]

Researchers have developed a technology that reconstructs the image of a transparent object hidden behind fog or an opaque barrier using only a single photograph.

KAIST announced Thursday that a research team led by Jang Mu-seok, a professor in the Department of Bio and Brain Engineering, has developed "single-shot phase imaging" technology capable of reconstructing a phase object completely concealed between two moving scattering layers from a single captured image.

A scattering layer is a medium — such as fog or an opaque film — that disperses light in multiple directions. A phase object is a transparent object, like glass, clear plastic wrap or a living cell, that is nearly indistinguishable from its surroundings under a conventional camera but causes subtle changes in light as it passes through.

By analyzing those minute phase shifts in light passing through such an object, researchers can determine its shape and optical thickness. Phase imaging technology is used to observe living cells without staining them and to inspect transparent components in semiconductors and displays.

However, when scattering layers in front of and behind an object move, the path of light changes continuously, making it difficult to accurately capture information about the object — much like trying to photograph something through a fogged window.

Existing techniques required multiple shots of the same subject or advance measurement of the scattering environment, and they needed AI models pre-trained on large datasets.

In response, the research team used a point-illumination method — which concentrates light into a single spot, much like a magnifying glass focusing sunlight — to ensure that light passing through the first scattering layer carried object information as stably as possible. The team then combined an optical model that calculates how light changes as it passes through the object and scattering layers with AI.

The technology first secures stable object information through point illumination, then uses the optical model and AI to trace the scattering process in reverse. The key achievement is that it reconstructs information about both the object and the scattering environment from a single image, without requiring multiple shots or prior measurement of the scattering conditions.

Using only the light-intensity data from a single shot, the team simultaneously determined the shape and thickness of a transparent object, the degree to which light had scattered, and the object's position.

The system operated reliably without additional training data even when the object's position changed or shifting scattering layers altered the degree of light distortion. In experiments across a range of scattering environments, it reconstructed object boundaries and fine structural details more clearly than existing single-shot phase recovery techniques.

The technology could be applied to the precise inspection of transparent components and optical materials used in semiconductors and displays that are difficult to examine with conventional optical equipment. It is also expected to find use in biotech imaging for observing living cells without damage and in next-generation optical sensors.

"This research marks the first case of restoring the shape and position of a transparent object from a single shot, even in environments where light is severely disrupted by fog or opaque barriers," Jang said. "We plan to develop the technology further so it operates reliably in even more complex environments, and to apply it across diverse fields including semiconductor inspection and biotech imaging."

Kim Yu-seon, a master's candidate, and Song Guk-ho, a doctoral candidate, both in KAIST's Department of Bio and Brain Engineering, served as co-first authors on the study, with Jang as the corresponding author. The findings were published July 20 in Optica, an international journal in the field of optics.

woo@heraldcorp.com