Far-field super-resolution ghost imaging with a deep neural network constraint

Light Science & Applications, Journal Year: 2022, Volume and Issue: 11(1)

Published: Jan. 1, 2022

Ghost imaging (GI) facilitates image acquisition under low-light conditions by single-pixel measurements and thus has great potential in applications various fields ranging from biomedical to remote sensing. However, GI usually requires a large amount of samplings order reconstruct high-resolution image, imposing practical limit for its applications. Here we propose far-field super-resolution technique that incorporates the physical model formation into deep neural network. The resulting hybrid network does not need pre-train on any dataset, allows reconstruction with resolution beyond diffraction limit. Furthermore, imposes constraint output, making it effectively interpretable. We experimentally demonstrate proposed flying drone, show outperforms some other widespread techniques terms both spatial sampling ratio. believe this study provides new framework GI, paves way

Language: Английский

---

Intensity-based holographic imaging via space-domain Kramers–Kronig relations

Nature Photonics, Journal Year: 2021, Volume and Issue: 15(5), P. 354 - 360

Published: Feb. 8, 2021

Language: Английский

---

On the use of deep learning for phase recovery

Light Science & Applications, Journal Year: 2024, Volume and Issue: 13(1)

Published: Jan. 1, 2024

Phase recovery (PR) refers to calculating the phase of light field from its intensity measurements. As exemplified quantitative imaging and coherent diffraction adaptive optics, PR is essential for reconstructing refractive index distribution or topography an object correcting aberration system. In recent years, deep learning (DL), often implemented through neural networks, has provided unprecedented support computational imaging, leading more efficient solutions various problems. this review, we first briefly introduce conventional methods PR. Then, review how DL provides following three stages, namely, pre-processing, in-processing, post-processing. We also used in image processing. Finally, summarize work provide outlook on better use improve reliability efficiency Furthermore, present a live-updating resource ( https://github.com/kqwang/phase-recovery ) readers learn about

Language: Английский

---

Artificial intelligence-enabled quantitative phase imaging methods for life sciences

Nature Methods, Journal Year: 2023, Volume and Issue: 20(11), P. 1645 - 1660

Published: Oct. 23, 2023

Language: Английский

---

Diffraction tomography with a deep image prior

Optics Express, Journal Year: 2020, Volume and Issue: 28(9), P. 12872 - 12872

Published: March 25, 2020

We present a tomographic imaging technique, termed Deep Prior Diffraction Tomography (DP-DT), to reconstruct the 3D refractive index (RI) of thick biological samples at high resolution from sequence low-resolution images collected under angularly varying illumination. DP-DT processes multi-angle data using phase retrieval algorithm that is extended by deep image prior (DIP), which reparameterizes sample reconstruction with an untrained, generative convolutional neural network (CNN). show effectively addresses missing cone problem, otherwise degrades and quality standard algorithms. As does not require pre-captured or pre-training, it biased towards any particular dataset. Hence, general technique can be applied wide variety samples, including scenarios in large datasets for supervised training would infeasible expensive. obtain RI maps bead phantoms complex specimens, both simulation experiment, produces higher-quality results than regularization techniques. further demonstrate generality DP-DT, two different scattering models, first Born multi-slice models. Our point potential benefits other modalities, X-ray computed tomography, magnetic resonance imaging, electron microscopy.

Language: Английский

---

Deep image prior for undersampling high-speed photoacoustic microscopy

Photoacoustics, Journal Year: 2021, Volume and Issue: 22, P. 100266 - 100266

Published: April 1, 2021

Photoacoustic microscopy (PAM) is an emerging imaging method combining light and sound. However, limited by the laser's repetition rate, state-of-the-art high-speed PAM technology often sacrifices spatial sampling density (i.e., undersampling) for increased speed over a large field-of-view. Deep learning (DL) methods have recently been used to improve sparsely sampled images; however, these require time-consuming pre-training training dataset with ground truth. Here, we propose use of deep image prior (DIP) quality undersampled images. Unlike other DL approaches, DIP requires neither nor fully-sampled truth, enabling its flexible fast implementation on various targets. Our results demonstrated substantial improvement in images as few 1.4$\%$ fully pixels PAM. approach outperforms interpolation, competitive pre-trained supervised method, readily translated high-speed, undersampling modalities.

Language: Английский

---

Untrained Neural Network Priors for Inverse Imaging Problems: A Survey

IEEE Transactions on Pattern Analysis and Machine Intelligence, Journal Year: 2022, Volume and Issue: unknown, P. 1 - 20

Published: Jan. 1, 2022

In recent years, advancements in machine learning (ML) techniques, particular, deep (DL) methods have gained a lot of momentum solving inverse imaging problems, often surpassing the performance provided by hand-crafted approaches. Traditionally, analytical been used to solve problems such as image restoration, inpainting, and superresolution. Unlike for which problem is explicitly defined domain knowledge carefully engineered into solution, DL models do not benefit from prior instead make use large datasets predict an unknown solution problem. Recently, new paradigm training using single image, named untrained neural network (UNNP) has proposed variety tasks, e.g., restoration inpainting. Since then, many researchers various applications variants UNNP. this paper, we present comprehensive review studies UNNP different tasks highlight open research require further research.

Language: Английский

---

DeepEIT: Deep Image Prior Enabled Electrical Impedance Tomography

IEEE Transactions on Pattern Analysis and Machine Intelligence, Journal Year: 2023, Volume and Issue: 45(8), P. 9627 - 9638

Published: Feb. 1, 2023

Neural networks (NNs) have been widely applied in tomographic imaging through data-driven training and image processing. One of the main challenges using NNs real medical is requirement massive amounts data which are not always available clinical practice. In this paper, we demonstrate that, on contrary, one can directly execute reconstruction without data. The key idea to bring recently introduced deep prior (DIP) merge it with electrical impedance tomography (EIT) reconstruction. DIP provides a novel approach regularization EIT problems by compelling recovered be synthesized from given NN architecture. Then, relying NN's built-in back-propagation, finite element solver, conductivity distribution optimized. Quantitative results based simulation experimental show that proposed method an effective unsupervised capable outperforming state-of-the-art alternatives.

Language: Английский

---

High-resolution, large field-of-view label-free imaging via aberration-corrected, closed-form complex field reconstruction

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: June 3, 2024

Abstract Computational imaging methods empower modern microscopes to produce high-resolution, large field-of-view, aberration-free images. Fourier ptychographic microscopy can increase the space-bandwidth product of conventional microscopy, but its iterative reconstruction are prone parameter selection and tend fail under excessive aberrations. Spatial Kramers–Kronig analytically reconstruct complex fields, is limited by aberration or providing extended resolution enhancement. Here, we present APIC, a closed-form method that weds strengths both while using only NA-matching darkfield measurements. We establish an analytical phase retrieval framework which demonstrates feasibility reconstructing field associated with APIC retrieve aberrations system no additional hardware avoids algorithms, requiring human-designed convergence metrics always obtaining solution. experimentally demonstrate gives correct results where fails when constrained same number achieves 2.8 times faster computation image tile size 256 (length-wise), robust against compared capable addressing whose maximal difference exceeds 3.8π NA 0.25 objective in experiment.

Language: Английский

---

Coordinate-based neural representations for computational adaptive optics in widefield microscopy

Nature Machine Intelligence, Journal Year: 2024, Volume and Issue: 6(6), P. 714 - 725

Published: June 24, 2024

Language: Английский

---