Parallel compressive super-resolution imaging with wide field-of-view based on physics enhanced network

TL;DR

This paper introduces a physics-enhanced neural network approach for wide field-of-view super-resolution imaging that achieves high-quality, real-time results with minimal masks, significantly advancing rapid imaging in various spectra.

Contribution

It proposes a novel physics-enhanced network trained with prior optical transfer functions for efficient, wide FOV super-resolution imaging using minimal masks and achieving real-time performance.

Findings

Achieves 4x4 super-resolution with only three masks.

Supports up to 1020x1500 wide FOV imaging.

Demonstrates effectiveness through simulations and experiments.

Abstract

Achieving both high-performance and wide field-of-view (FOV) super-resolution imaging has been attracting increasing attention in recent years. However, such goal suffers from long reconstruction time and huge storage space. Parallel compressive imaging (PCI) provides an efficient solution, but the super-resolution quality and imaging speed are strongly dependent on precise optical transfer function (OTF), modulation masks and reconstruction algorithm. In this work, we propose a wide FOV parallel compressive super-resolution imaging approach based on physics enhanced network. By training the network with the prior OTF of an arbitrary 128x128-pixel region and fine-tuning the network with other OTFs within rest regions of FOV, we realize both mask optimization and super-resolution imaging with up to 1020x1500 wide FOV. Numerical simulations and practical experiments demonstrate the effectiveness and superiority of the proposed approach. We achieve high-quality reconstruction with 4x4 times super-resolution enhancement using only three designed masks to reach real-time imaging speed. The proposed approach promotes the technology of rapid imaging for super-resolution and wide FOV, ranging from infrared to Terahertz.