Metasurfaces perform all-optical image edge detection

Metasurfaces perform all-optical image edge detection

Technology News |
Leveraging the special optical properties of metasurfaces to revisit analog optical computing, researchers from AMOLF (institute for Atomic and Molecular Physics) together with their collaborators from the Advanced Science Research Center (ASRC/CUNY) have designed a nanostructured surface capable of performing all-optical edge detection on an input image.
By eeNews Europe


Being passive devices, such “computing metasurfaces” perform the mathematical operations without requiring any power, hence they could dramatically slash the power consumption of object recognition systems where edge detection is a pre-requisite. In a paper titled “High-Index Dielectric Metasurfaces Performing Mathematical Operations” published in the Nano Letters, the researchers demonstrate optical metasurfaces composed of dielectric nanobeams, illuminated by light polarized along the beams’ direction.

(left) Schematic of edge detection and spatial differentiation;
(right) derivative image of AMOLF logo taken at a wavelength
of 726nm. Credit: AMOLF

When an image is projected onto the transparent metasurface, the transmitted light forms a new image that shows the edges of the original. Effectively, the metasurface performs a mathematical derivative operation on the image which provides a direct probe of edges in the image.

Typically, this edge detection is performed in the digital domain on images captured by a sensor, requiring extra computing steps and implying fundamental speed limitations and high energy consumption. Here, the optical mathematical transformation results from the fact that each spatial frequency that composes the image has a tailored transmission coefficient through the metasurface. This tailored transmission is the result of a complex interference of light as it propagates through the metasurface.

The authors anticipate that such transparent computing metasurfaces could be integrated at sensor level, directly onto a standard CCD or CMOS detector chip to boost the speed of existing imaging processing techniques and lower energy usage, for ultrafast object detection in augmented reality applications.


Related articles:

Researchers squeeze laser light into quantum processor

Startup reveals prototype optical AI processor

Artificial neural networks go all-optical

Linked Articles
eeNews Europe