4K-DMDNet: diffraction model-driven network for 4K computer-generated holography

Opto-Electronic Advances, Journal Year: 2023, Volume and Issue: 6(5), P. 220135 - 220135

Published: Jan. 1, 2023

Deep learning offers a novel opportunity to achieve both high-quality and high-speed computer-generated holography (CGH). Current data-driven deep algorithms face the challenge that labeled training datasets limit performance generalization. The model-driven introduces diffraction model into neural network. It eliminates need for dataset has been extensively applied hologram generation. However, existing problem of insufficient constraints. In this study, we propose network capable high-fidelity 4K generation, called Diffraction Model-driven Network (4K-DMDNet). constraint reconstructed images in frequency domain is strengthened. And structure combines residual method sub-pixel convolution built, which effectively enhances fitting ability inverse problems. generalization 4K-DMDNet demonstrated with binary, grayscale 3D images. High-quality full-color optical reconstructions holograms have achieved at wavelengths 450 nm, 520 638 nm.

Language: Английский

---

Fourier Imager Network (FIN): A deep neural network for hologram reconstruction with superior external generalization

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

Published: Aug. 16, 2022

Deep learning-based image reconstruction methods have achieved remarkable success in phase recovery and holographic imaging. However, the generalization of their performance to new types samples never seen by network remains a challenge. Here we introduce deep learning framework, termed Fourier Imager Network (FIN), that can perform end-to-end from raw holograms samples, exhibiting unprecedented external generalization. FIN architecture is based on spatial transform modules process frequencies its inputs using learnable filters global receptive field. Compared with existing convolutional neural networks used for hologram reconstruction, exhibits superior while also being much faster inference speed, completing task ~0.04 s per 1 mm2 sample area. We experimentally validated training it human lung tissue blindly testing prostate, salivary gland Pap smear proving speed. Beyond microscopy quantitative imaging, underlying might open up various opportunities design broadly generalizable models computational imaging machine vision fields.

Language: Английский

---

Self-supervised learning of hologram reconstruction using physics consistency

Nature Machine Intelligence, Journal Year: 2023, Volume and Issue: 5(8), P. 895 - 907

Published: Aug. 7, 2023

Abstract Existing applications of deep learning in computational imaging and microscopy mostly depend on supervised learning, requiring large-scale, diverse labelled training data. The acquisition preparation such image datasets is often laborious costly, leading to limited generalization new sample types. Here we report a self-supervised model, termed GedankenNet, that eliminates the need for or experimental data, demonstrate its effectiveness superior hologram reconstruction tasks. Without prior knowledge about types, model was trained using physics-consistency loss artificial random images synthetically generated without any experiments resemblance real-world samples. After training, GedankenNet successfully generalized holograms unseen biological samples, reconstructing phase amplitude different types object experimentally acquired holograms. access real samples their spatial features, achieved complex-valued reconstructions consistent with wave equation free space. framework also shows resilience random, unknown perturbations physical forward including changes distances, pixel size illumination wavelength. This creates opportunities solving inverse problems holography, imaging.

Language: Английский

---

Toward immersive communications in 6G

Frontiers in Computer Science, Journal Year: 2023, Volume and Issue: 4

Published: Jan. 11, 2023

The sixth generation (6G) networks are expected to enable immersive communications and bridge the physical virtual worlds. Integrating extended reality, holography, haptics, will revolutionize how people work, entertain, communicate by enabling lifelike interactions. However, unprecedented demand for data transmission rate stringent requirements on latency reliability create challenges 6G support communications. In this survey article, we present prospect of investigate emerging solutions corresponding 6G. First, introduce use cases communications, in fields entertainment, education, healthcare. Second, concepts including haptic communication, holographic their basic implementation procedures, terms rate, latency, reliability. Third, summarize potential addressing from aspects computing, networking. Finally, discuss future research directions conclude study.

Language: Английский

---

Non-convex optimization for inverse problem solving in computer-generated holography

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

Published: July 9, 2024

Abstract Computer-generated holography is a promising technique that modulates user-defined wavefronts with digital holograms. Computing appropriate holograms faithful reconstructions not only problem closely related to the fundamental basis of but also long-standing challenge for researchers in general fields optics. Finding exact solution desired hologram reconstruct an accurate target object constitutes ill-posed inverse problem. The practice single-diffraction computation synthesizing can provide approximate answer, which subject limitations numerical implementation. Various non-convex optimization algorithms are thus designed seek optimal by introducing different constraints, frameworks, and initializations. Herein, we overview applied computer-generated holography, incorporating principles synthesis based on alternative projections gradient descent methods. This aimed underlying optimized generation, as well insights into cutting-edge developments this rapidly evolving field potential applications virtual reality, augmented head-up display, data encryption, laser fabrication, metasurface design.

Language: Английский

---

Thermally tunable binary-phase VO₂ metasurfaces for switchable holography and digital encryption

Nanophotonics, Journal Year: 2024, Volume and Issue: 13(7), P. 1109 - 1117

Published: Feb. 21, 2024

Metasurface holography has aroused immense interest in producing holographic images with high quality, higher-order diffraction-free, and large viewing angles by using a planar artificial sheet consisting of subwavelength nanostructures. Despite remarkable progress, dynamically tunable metasurface the visible band rarely been reported due to limited available tuning methods. In this work, we propose numerically demonstrate thermally vanadium dioxide (VO

Language: Английский

---

Roadmap on computational methods in optical imaging and holography [invited]

Applied Physics B, Journal Year: 2024, Volume and Issue: 130(9)

Published: Aug. 29, 2024

Computational methods have been established as cornerstones in optical imaging and holography recent years. Every year, the dependence of on computational is increasing significantly to extent that components are being completely efficiently replaced with at low cost. This roadmap reviews current scenario four major areas namely incoherent digital holography, quantitative phase imaging, through scattering layers, super-resolution imaging. In addition registering perspectives modern-day architects above research areas, also reports some latest studies topic. codes pseudocodes presented for a plug-and-play fashion readers not only read understand but practice algorithms their data. We believe this will be valuable tool analyzing trends predict prepare future holography.

Language: Английский

---

DH-GAN: a physics-driven untrained generative adversarial network for holographic imaging

Optics Express, Journal Year: 2023, Volume and Issue: 31(6), P. 10114 - 10114

Published: Feb. 6, 2023

Digital holography is a 3D imaging technique by emitting laser beam with plane wavefront to an object and measuring the intensity of diffracted waveform, called holograms. The object's shape can be obtained numerical analysis captured holograms recovering incurred phase. Recently, deep learning (DL) methods have been used for more accurate holographic processing. However, most supervised require large datasets train model, which rarely available in DH applications due scarcity samples or privacy concerns. A few one-shot DL-based recovery exist no reliance on paired images. Still, these often neglect underlying physics law that governs wave propagation. These offer black-box operation, not explainable, generalizable, transferrable other applications. In this work, we propose new DL architecture based generative adversarial networks uses discriminative network realizing semantic measure reconstruction quality while using as function approximator model inverse hologram formation. We impose smoothness background part recovered image progressive masking module powered simulated annealing enhance quality. proposed method exhibits high transferability similar samples, facilitates its fast deployment time-sensitive without need retraining from scratch. results show considerable improvement competitor (about 5 dB PSNR gain) robustness noise 50% reduction vs increase rate).

Language: Английский

---

eFIN: Enhanced Fourier Imager Network for Generalizable Autofocusing and Pixel Super-Resolution in Holographic Imaging

IEEE Journal of Selected Topics in Quantum Electronics, Journal Year: 2023, Volume and Issue: 29(4: Biophotonics), P. 1 - 10

Published: Feb. 24, 2023

The application of deep learning techniques has greatly enhanced holographic imaging capabilities, leading to improved phase recovery and image reconstruction. Here, we introduce a neural network termed Fourier Imager Network (eFIN) as highly generalizable robust framework for hologram reconstruction with pixel super-resolution autofocusing. Through microscopy experiments involving lung, prostate salivary gland tissue sections Papanicolau (Pap) smears, demonstrate that eFIN superior quality exhibits external generalization new types samples never seen during the training phase. This achieves wide autofocusing axial range Δz ∼ 350 μm, capability accurately predict distances by physics-informed learning. enables 3× increases space-bandwidth product reconstructed images 9-fold almost no performance loss, which allows significant time savings in data processing steps. Our results showcase advancements pushing boundaries various applications e.g., quantitative label-free microscopy.

Language: Английский

---

Multi-depth hologram generation from two-dimensional images by deep learning

Optics and Lasers in Engineering, Journal Year: 2023, Volume and Issue: 170, P. 107758 - 107758

Published: Aug. 2, 2023

Language: Английский

---