Face Recognition Example Under Drastic Brightness Variations . (Image courtesy of the National Research Foundation of Korea)
DAEJEON, Aug. 19 (Korea Bizwire) – A South Korean research team has developed a novel AI-powered image sensor inspired by the human brain’s neural architecture, capable of recognizing visual information under extreme lighting conditions without the need for retraining or additional data correction.
According to the National Research Foundation of Korea, the joint research team, led by Professor Song Young-min of KAIST and Professor Kang Dong-ho of GIST, unveiled a ferroelectric-based optical sensor that mimics synaptic interaction between neurons and glial cells—an essential dynamic for human learning and perception.
Conventional CMOS image sensors often struggle with rapid brightness changes—such as moving from indoor to outdoor environments—resulting in data loss due to overexposure. These sensors typically require complex post-processing to compensate.
The team’s breakthrough draws on the polarization-retaining properties of ferroelectric materials to create a light-sensitive device that can store, amplify, or suppress visual information. This architecture, modeled after brain-like neuromorphic structures, allows the sensor to handle drastic lighting variations in real time, such as detecting faces in low light or overly bright environments, without retraining.
“This study marks a significant expansion of ferroelectric devices beyond their conventional role in memory storage, applying them to neuromorphic AI sensing,” said Professor Song. “It has promising implications for applications in autonomous vehicles, smart robotics, and advanced AI vision systems.”
The research findings were published in the July 28 online edition of the international journal Advanced Materials.
Kevin Lee (kevinlee@koreabizwire.com)
