On a chilly, sunny day, you’re driving on a rural highway, surrounded by snow-covered fields. Right away, your eyes course of the scene, choosing out particular person objects to concentrate on — a cease signal, a barn — whereas the remainder of the scene blurs within the periphery. Your mind shops the targeted and blurred photographs as a reminiscence that may be pictured in your thoughts later, whereas sitting at your desk.
Mimicking this simple, instantaneous picture processing energy of the human eye, Penn State electrical engineering researchers created a metasurface: an optical component akin to a glass slide that makes use of tiny nanostructures positioned at completely different angles to manage mild. Led by corresponding writer Xingjie Ni, affiliate professor {of electrical} engineering and pc science (EECS) at Penn State, the group revealed their invention in Nature Communications.
Synthetic intelligence (AI) methods require important computing energy and vitality and may be sluggish to course of photographs and establish objects, in accordance with the researchers. Against this, the metasurface can be utilized to preprocess and rework photographs earlier than they’re captured by a digicam, permitting a pc — and AI — to course of them with minimal energy and knowledge bandwidth.
The metasurface works by changing a picture from the Cartesian coordinate system, the place picture pixels are organized in straight rows and columns alongside the x and y axes, to the log-polar system, which makes use of a bullseye-like pixel distribution.
“Just like the association of sunshine receptors contained in the human eye, the metasurface takes photographs and arranges them in a log-polar coordinate system — with denser pixels for the central, targeted options and sparser pixels for the peripheral areas,” Ni mentioned. “This enables for the extra necessary features of a photograph to return by clearly whereas others stay much less in focus, thereby saving knowledge bandwidth.”
The metasurface is positioned in entrance of a digicam in order that mild first passes by it and transforms the picture from the Cartesian system into log-polar coordinates earlier than it’s digitalized by a digicam and transferred to a pc. Since it really works utilizing nanostructures that bend mild, the metasurface doesn’t want any energy and works on the pace of sunshine.
“As a picture of an object can fluctuate in measurement or orientation, it’s fascinating to preprocess photographs to make them proof against scale and rotation adjustments,” Ni mentioned. “This preprocessing helps AI functions extra simply acknowledge them as the identical object.”
By putting a distinct metasurface in entrance of a digicam, researchers can also rework the log-polar picture again into the unique picture with Cartesian coordinates.
The invention has many potential functions, the researchers mentioned, together with to be used in goal monitoring and surveillance to map how a automobile, for instance, strikes throughout a metropolis.
“A metasurface can be utilized in tandem with AI methods as a preprocessor, making it simpler to acknowledge the identical automobile from a number of road view cameras,” Ni mentioned. “Or whether it is utilized to a satellite tv for pc, it might doubtlessly observe planes from takeoff to touchdown.”
Unique Article: Optical invention mirrors the image processing power of a human eye
Extra from: Pennsylvania State University
