SEOUL, August 06 (AJP) - A KAIST research team has developed an imaging method that finds a transparent object buried inside fog and works out its shape, thickness and position from a single camera exposure, the university said Thursday.
Glass, clear plastic film and living cells are almost invisible to an ordinary camera because they barely change the brightness of the light passing through them. They do change something else. Light slows very slightly as it crosses them, and that shift, called phase, carries the object's outline and its optical thickness. Reading it is how scientists inspect transparent parts in semiconductors and displays, or watch living cells without staining them.
The reading breaks down when something scatters the light on the way. Fog, frosted glass and cloudy tissue all send light off in random directions, and when those layers are moving, the scrambling changes moment to moment. Until now the workarounds were to take many exposures of the same target, to measure the scattering environment in advance, or to train a system on large volumes of example images.
The team led by Jang Moo-seok, a professor in KAIST's Department of Bio and Brain Engineering, removed all three requirements. Their technique recovers an object sealed between two moving scattering layers, with fog in front of it and fog behind it, from one shot.
Two ideas make that possible. The first is where the light goes in. Instead of lighting up the whole front layer, the researchers focus the beam onto a very small spot, the way a magnifying glass concentrates sunlight. Fewer paths through the layer means the light stays organized enough to record the object's phase in the pattern that reaches the camera.
The second is where the camera sits. Placed close to the far side of the rear layer, it does not receive a purely random mess. It receives the object's pattern with a consistent smear laid over it, which the team modeled as scattering blur. A random problem became a solvable one.
From there the method uses a physical model of how light travels through the object and the fog, paired with artificial intelligence. It does not work the way image AI usually does, by studying thousands of correct answers first. It runs the physics backward from the single measurement, testing what object, what blur and what distance would together produce exactly the pattern the camera saw.
All three answers come out at once. The technique held up when the object moved and when the scattering changed, with no calibration and no training data, and in testing it reproduced edges and fine structure more sharply than existing single-shot phase methods.
"This is the first demonstration of restoring the shape and position of a transparent object from a single measurement, even in environments where light is severely scattered, such as behind fog or a diffusive film," Jang said. He said the team plans to extend the method to more complex environments, with semiconductor inspection and biomedical imaging as targets.
KAIST listed three directions for the work. In manufacturing, it could inspect transparent components sitting under scattering coatings or protective films. In medicine, it could image the retina through a cataract. Applied to reflected rather than transmitted light, it could image objects out of direct line of sight, including for military use.
Kim Yoo-sun, a master's student, and Song Gook-ho, a doctoral candidate, were co-first authors, with Jang as corresponding author. The paper appeared in the optics journal Optica on July 20, and the National Research Foundation of Korea funded the work under the Ministry of Science and ICT.
Copyright ⓒ Aju Press All rights reserved.