Metalenses and neural array allow compact high-quality imaging device

03 Dec 2025

Tongji University platform balances trade-off between aperture, bandwidth, image quality.

A project group including Tongji University, Stanford University and Shanghai Institute of Technical Physics has developed a camera said to help solve the inherent imaging trade-off between bandwidth and image quality.

Described in eLight the new principle involves a combination of metalenses and computational imaging, and could prove valuable for scientific, industrial, and consumer applications.

"Demand for compact, high-quality, video-rate, full-color imaging systems is indispensable in modern society," noted the project in its paper.

"However, progress toward such systems is hindered by the bulky form factor of conventional refractive optics."

Metasurface lenses and their non-traditional optical behavior are a valuable route to new imaging architectures taking up smaller volumes than traditional platforms. But metalenses can still suffer from significant chromatic aberration and constraints on the aperture sizes than can be employed.

Computational methods can help shift part of the aberration correction from optics to algorithms, although no existing work has successfully achieved this goal, commented the project.

One problem stems from the modulation transfer function (MTF), a parameter relating to the ability of a lens to transfer contrast at a particular resolution from the object to the image - the information transfer capability of the imager. Under normal circumstances, even computational image reconstruction becomes inaccurate when the MTF falls below a noise-dependent threshold.

Route to thinner more compact cameras

The project's solution is an imaging system combining a metalens array, an RGB CMOS sensor and a computing layer chip. Crucially the metalens array "leverages an innovative neural array imaging model to capture wide field-of-view and broadband light information, which the CMOS sensor records as multiplexed measurements."

In a neural array model, a number of discrete computational units share data processing so they can act as a single AI engine. A further deconvolution algorithm then reconstructs high-fidelity images of a target.

"A key challenge lies in determining the optimal number and arrangement of small-aperture lenses," wrote the project. "Simple periodic layouts introduce multiple zero-frequency points in the MTF, whereas breaking periodicity effectively avoids this issue."

To test its platform, the project built a neural array imaging prototype with an aperture of 2.76 millimeters, 50° field of view and spectral range of 400 to 700 nanometers. This set-up matched the MTF of a commercial compound lens and imaged successfully in both indoor and outdoor environments. The total optical track length was greatly reduced, from 57 to 4.3 millimeters, a 13-fold reduction in thickness.

In real world scenarios the neural array imaging system was used for depth estimation and object detection, both vital factors for use in autonomous navigation, machine vision, surveillance and augmented reality applications.

"The neural array mapping technique is not limited to metasurfaces, but also applicable to traditional refractive lens systems," noted the project group. "This approach provides a promising route for miniaturizing imaging systems, paving the way for thinner, more compact cameras in the future."