Zhengyu Huang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Zhengyu Huang has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
Theodore B. Norris
University of Michigan-Ann Arbor
4 shared publications- Jeffrey A. Fessler
University of Michigan
3 shared publications - Shaobin Liu
Nanjing University of Aeronautics and Astronautics
3 shared publications - Feng Xue
Nanjing University of Aeronautics and Astronautics,College of Information Science and Technology,Nanjing,China,210016
2 shared publications - Il Yong Chun
Department of Electrical and Computer Engineering, Sungkyunkwan University, Suwon, Republic of Korea
2 shared publications - Lingling Wang
School of Electronic and Information Engineering, Jinling Institute of Technology, Nanjing, China
2 shared publications
Latest Publications
- 8 September 2022
- Published by Optica Publishing Group in Optics Continuum
- Vol. 1 (9) , 2030-2042
Focal stack camera: depth estimation performance comparison and design exploration
The concept of focal stack cameras based on transparent sensors has been recently demonstrated, enabling depth sensing in a single exposure from a focal stack. This study investigates how the depth estimation accuracy is affected by numerous camera parameters, including aperture size, sensor resolution and number of sensor planes. We investigate the dependence of these parameters on the deep learning based depth estimation performance and make a comparison between the focal stack camera and the light field camera, thus providing guidelines for future focal stack camera design.
- 8 September 2022
- Published by Optica Publishing Group in Optics Continuum
- Vol. 1 (9) , 2030-2042
Focal stack camera: depth estimation performance comparison and design exploration
The concept of focal stack cameras based on transparent sensors has been recently demonstrated, enabling depth sensing in a single exposure from a focal stack. This study investigates how the depth estimation accuracy is affected by numerous camera parameters, including aperture size, sensor resolution and number of sensor planes. We investigate the dependence of these parameters on the deep learning based depth estimation performance and make a comparison between the focal stack camera and the light field camera, thus providing guidelines for future focal stack camera design.
- Open Access
Unsupervised Learning Based Focal Stack Camera Depth Estimation
We propose an unsupervised deep learning based method to estimate depth from focal stack camera images. On the NYU-v2 dataset, our method achieves much better depth estimation accuracy compared to single-image based methods.
Open AccessUnsupervised Learning Based Focal Stack Camera Depth Estimation
We propose an unsupervised deep learning based method to estimate depth from focal stack camera images. On the NYU-v2 dataset, our method achieves much better depth estimation accuracy compared to single-image based methods.
- 1 January 2022
- Published by Optica Publishing Group
- p. JW3A.5
Unsupervised Learning Based Focal Stack Camera Depth Estimation
We propose an unsupervised deep learning based method to estimate depth from focal stack camera images. On the NYU-v2 dataset, our method achieves much better depth estimation accuracy compared to single-image based methods.
- 1 January 2022
- Published by Optica Publishing Group
- p. JW3A.5
Unsupervised Learning Based Focal Stack Camera Depth Estimation
We propose an unsupervised deep learning based method to estimate depth from focal stack camera images. On the NYU-v2 dataset, our method achieves much better depth estimation accuracy compared to single-image based methods.
- 1 July 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 19-22
One-shot Line Extraction from Color Illustrations
Sketch colorization has been explored extensively using deep learning-based approaches. However, there are few works on the inverse problem of extracting lines from color illustrations, which is challenging due to the abstract illustration in cartoon rendering and the diversity of colorization style. In this paper, we propose a learning-based framework to explore one-shot line extraction from color illustrations. To avoid over-fitting, we simplified the proposed model and proposed a method for data augmentation when training using a single “color illustration-line drawing” pair. Then, we verified the model's effectiveness and efficiency in a group of color illustrations with similar colorization style. The experiments showed that the proposed, our method can outperform the conventional CNN (convolution neural network)-based sketch extraction methods.
- 1 July 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 19-22
One-shot Line Extraction from Color Illustrations
Sketch colorization has been explored extensively using deep learning-based approaches. However, there are few works on the inverse problem of extracting lines from color illustrations, which is challenging due to the abstract illustration in cartoon rendering and the diversity of colorization style. In this paper, we propose a learning-based framework to explore one-shot line extraction from color illustrations. To avoid over-fitting, we simplified the proposed model and proposed a method for data augmentation when training using a single “color illustration-line drawing” pair. Then, we verified the model's effectiveness and efficiency in a group of color illustrations with similar colorization style. The experiments showed that the proposed, our method can outperform the conventional CNN (convolution neural network)-based sketch extraction methods.
- 29 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 45 (4) , 4915-4931
Momentum-Net: Fast and Convergent Iterative Neural Network for Inverse Problems
Iterative neural networks (INN) are rapidly gaining attention for solving inverse problems in imaging, image processing, and computer vision. INNs combine regression NNs and an iterative model-based image reconstruction (MBIR) algorithm, often leading to both good generalization capability and outperforming reconstruction quality over existing MBIR optimization models. This paper proposes the first fast and convergent INN architecture, Momentum-Net , by generalizing a block-wise MBIR algorithm that uses momentum and majorizers with regression NNs. For fast MBIR, Momentum-Net uses momentum terms in extrapolation modules, and noniterative MBIR modules at each iteration by using majorizers , where each iteration of Momentum-Net consists of three core modules: image refining, extrapolation, and MBIR. Momentum-Net guarantees convergence to a fixed-point for general differentiable (non)convex MBIR functions (or data-fit terms) and convex feasible sets, under two asymptomatic conditions. To consider data-fit variations across training and testing samples, we also propose a regularization parameter selection scheme based on the “spectral spread” of majorization matrices. Numerical experiments for light-field photography using a focal stack and sparse-view computational tomography demonstrate that, given identical regression NN architectures, Momentum-Net significantly improves MBIR speed and accuracy over several existing INNs; it significantly improves reconstruction quality compared to a state-of-the-art MBIR method in each application.
- 29 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 45 (4) , 4915-4931
Momentum-Net: Fast and Convergent Iterative Neural Network for Inverse Problems
Iterative neural networks (INN) are rapidly gaining attention for solving inverse problems in imaging, image processing, and computer vision. INNs combine regression NNs and an iterative model-based image reconstruction (MBIR) algorithm, often leading to both good generalization capability and outperforming reconstruction quality over existing MBIR optimization models. This paper proposes the first fast and convergent INN architecture, Momentum-Net , by generalizing a block-wise MBIR algorithm that uses momentum and majorizers with regression NNs. For fast MBIR, Momentum-Net uses momentum terms in extrapolation modules, and noniterative MBIR modules at each iteration by using majorizers , where each iteration of Momentum-Net consists of three core modules: image refining, extrapolation, and MBIR. Momentum-Net guarantees convergence to a fixed-point for general differentiable (non)convex MBIR functions (or data-fit terms) and convex feasible sets, under two asymptomatic conditions. To consider data-fit variations across training and testing samples, we also propose a regularization parameter selection scheme based on the “spectral spread” of majorization matrices. Numerical experiments for light-field photography using a focal stack and sparse-view computational tomography demonstrate that, given identical regression NN architectures, Momentum-Net significantly improves MBIR speed and accuracy over several existing INNs; it significantly improves reconstruction quality compared to a state-of-the-art MBIR method in each application.
- Open Access
Momentum-Net: Fast and convergent iterative neural network for inverse problems
Iterative neural networks (INN) are rapidly gaining attention for solving inverse problems in imaging, image processing, and computer vision. INNs combine regression NNs and an iterative model-based image reconstruction (MBIR) algorithm, often leading to both good generalization capability and outperforming reconstruction quality over existing MBIR optimization models. This paper proposes the first fast and convergent INN architecture, Momentum-Net, by generalizing a block-wise MBIR algorithm that uses momentum and majorizers with regression NNs. For fast MBIR, Momentum-Net uses momentum terms in extrapolation modules, and noniterative MBIR modules at each iteration by using majorizers, where each iteration of Momentum-Net consists of three core modules: image refining, extrapolation, and MBIR. Momentum-Net guarantees convergence to a fixed-point for general differentiable (non)convex MBIR functions (or data-fit terms) and convex feasible sets, under two asymptomatic conditions. To consider data-fit variations across training and testing samples, we also propose a regularization parameter selection scheme based on the "spectral spread" of majorization matrices. Numerical experiments for light-field photography using a focal stack and sparse-view computational tomography demonstrate that, given identical regression NN architectures, Momentum-Net significantly improves MBIR speed and accuracy over several existing INNs; it significantly improves reconstruction quality compared to a state-of-the-art MBIR method in each application.
Open AccessMomentum-Net: Fast and convergent iterative neural network for inverse problems
Iterative neural networks (INN) are rapidly gaining attention for solving inverse problems in imaging, image processing, and computer vision. INNs combine regression NNs and an iterative model-based image reconstruction (MBIR) algorithm, often leading to both good generalization capability and outperforming reconstruction quality over existing MBIR optimization models. This paper proposes the first fast and convergent INN architecture, Momentum-Net, by generalizing a block-wise MBIR algorithm that uses momentum and majorizers with regression NNs. For fast MBIR, Momentum-Net uses momentum terms in extrapolation modules, and noniterative MBIR modules at each iteration by using majorizers, where each iteration of Momentum-Net consists of three core modules: image refining, extrapolation, and MBIR. Momentum-Net guarantees convergence to a fixed-point for general differentiable (non)convex MBIR functions (or data-fit terms) and convex feasible sets, under two asymptomatic conditions. To consider data-fit variations across training and testing samples, we also propose a regularization parameter selection scheme based on the "spectral spread" of majorization matrices. Numerical experiments for light-field photography using a focal stack and sparse-view computational tomography demonstrate that, given identical regression NN architectures, Momentum-Net significantly improves MBIR speed and accuracy over several existing INNs; it significantly improves reconstruction quality compared to a state-of-the-art MBIR method in each application.
- 1 December 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 312, 750-753
Plasmonic Band-pass/stop Filters Based on Metal-Insulator-Metal Slit Waveguides
A novel plasmonic band-pass/stop filter based on metal-insulator-metal (MIM) plasmonic waveguides in telecommunication systems is proposed and investigated. In order to study the propagation of these surface waves and formation of plasmonic modes in the gap distance between metals, we use the finite element method (FEM) as numerical methods and microwave Transmission Line Method (TLM) for analytical one. The results of these calculations shows a significant transmission/reflection around λ ~ 1.57 μm where these spectral responses are acceptable for realizing band-pass/stop filtering in telecommunication window. Changing the gap distance between metallic sides in MIM slit, leads to a red or blue shift in the filtering spectra. These results would be helpful in designing the band-stop filters and band-pass filters, and have guiding significance for plasmonic integration circuits applications.
- 1 December 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 312, 750-753
Plasmonic Band-pass/stop Filters Based on Metal-Insulator-Metal Slit Waveguides
A novel plasmonic band-pass/stop filter based on metal-insulator-metal (MIM) plasmonic waveguides in telecommunication systems is proposed and investigated. In order to study the propagation of these surface waves and formation of plasmonic modes in the gap distance between metals, we use the finite element method (FEM) as numerical methods and microwave Transmission Line Method (TLM) for analytical one. The results of these calculations shows a significant transmission/reflection around λ ~ 1.57 μm where these spectral responses are acceptable for realizing band-pass/stop filtering in telecommunication window. Changing the gap distance between metallic sides in MIM slit, leads to a red or blue shift in the filtering spectra. These results would be helpful in designing the band-stop filters and band-pass filters, and have guiding significance for plasmonic integration circuits applications.
- 1 December 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1645-1649
A Wideband Fabry-Perot Antenna with Low-RCS High-GBP Using Embed-chessboard Polarization Conversion Metasurface
In this paper, an Fabry-Perot (FP) antenna with high gain-bandwidth product (GBP) and wideband Radar Cross Section (RCS) reduction using embed-chessboard polarization conversion metasurface (embed-CPCM) is proposed. Two kinds of symmetrical polarization conversion units are etched on both sides of the dielectric substrate with chessboard arranged to form the embed-CPCM. Wideband low-RCS feature of the proposed FP antenna is realized by phase cancellation, and gain enhancement is achieved by FP cavity. The simulated results shows that the FP antenna with embed-CPCM has the 10-dB impedance bandwidth of 8.55–12.11 GHz (34.5%), the 3-dB gain bandwidth of 8.95–11.54 GHz (25.3%), and the maximum realized gain is 17.31 dBi at 10.2 GHz, and the RCS reduction band is 8–26 GHz with peak RCS reduction of 21.06 dB at 9.7 GHz.
- 1 December 2019
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1645-1649
A Wideband Fabry-Perot Antenna with Low-RCS High-GBP Using Embed-chessboard Polarization Conversion Metasurface
In this paper, an Fabry-Perot (FP) antenna with high gain-bandwidth product (GBP) and wideband Radar Cross Section (RCS) reduction using embed-chessboard polarization conversion metasurface (embed-CPCM) is proposed. Two kinds of symmetrical polarization conversion units are etched on both sides of the dielectric substrate with chessboard arranged to form the embed-CPCM. Wideband low-RCS feature of the proposed FP antenna is realized by phase cancellation, and gain enhancement is achieved by FP cavity. The simulated results shows that the FP antenna with embed-CPCM has the 10-dB impedance bandwidth of 8.55–12.11 GHz (34.5%), the 3-dB gain bandwidth of 8.95–11.54 GHz (25.3%), and the maximum realized gain is 17.31 dBi at 10.2 GHz, and the RCS reduction band is 8–26 GHz with peak RCS reduction of 21.06 dB at 9.7 GHz.
- Open Access
Terahertz Circular Dichroism Spectroscopy of Biomaterials Enabled by Kirigami Polarization Modulators
S: Terahertz circular dichroism (TCD) offers spectroscopic capabilities for understanding mesoscale chiral architecture and low-energy vibrations of macromolecules in (bio)materials 1–5 . However, the lack of dynamic polarization modulators comparable to polarization optics for other parts of electromagnetic spectrum impedes proliferation of TCD spectroscopy 6–10 . Here we show that tunable optical elements fabricated from patterned plasmonic sheets with periodic kirigami cuts make possible polarization modulation of THz radiation under application of mechanical strain. A herringbone pattern of microscale metal stripes enables dynamic range of polarization rotation modulation exceeding 80° over thousands of cycles. Upon out-of-plane buckling, the plasmonic stripes function as reconfigurable semi-helices of variable pitch aligned along the THz propagation direction. Several biomaterials, exemplified by elytrons of Chrysina gloriosa beetles, revealed distinct TCD fingerprints associated with the helical substructure in the biocomposite. Analogous kirigami modulators will also enable other applications in THz optics, such as polarization-based terahertz imaging and phase-encrypted telecommunication.
Open AccessTerahertz Circular Dichroism Spectroscopy of Biomaterials Enabled by Kirigami Polarization Modulators
S: Terahertz circular dichroism (TCD) offers spectroscopic capabilities for understanding mesoscale chiral architecture and low-energy vibrations of macromolecules in (bio)materials 1–5 . However, the lack of dynamic polarization modulators comparable to polarization optics for other parts of electromagnetic spectrum impedes proliferation of TCD spectroscopy 6–10 . Here we show that tunable optical elements fabricated from patterned plasmonic sheets with periodic kirigami cuts make possible polarization modulation of THz radiation under application of mechanical strain. A herringbone pattern of microscale metal stripes enables dynamic range of polarization rotation modulation exceeding 80° over thousands of cycles. Upon out-of-plane buckling, the plasmonic stripes function as reconfigurable semi-helices of variable pitch aligned along the THz propagation direction. Several biomaterials, exemplified by elytrons of Chrysina gloriosa beetles, revealed distinct TCD fingerprints associated with the helical substructure in the biocomposite. Analogous kirigami modulators will also enable other applications in THz optics, such as polarization-based terahertz imaging and phase-encrypted telecommunication.
- Open Access
- 1 February 2019
- Published by AIP Publishing in APL Photonics
- Vol. 4 (2) , 026103
Nanoscale fingerprinting with hyperbolic metamaterials
We develop a new concept for optical sub-wavelength discrimination of nanostructures positioned on a hyperbolic metamaterial (HMM) surface using only far-field measurements of the spectral scattered intensity. The device demonstrating our approach consists of an HMM structure, one strong scatterer on top of the HMM at an a priori known location, and the target nanostructures to be sensed on the opposite interface of the HMM. Under far-field illumination from the top, highly localized beams are generated within the HMM by the top scatterer, with wavelength-dependent propagation directions. If the incident wavelength is varied, the localized beam will sweep across the target(s) and the far-field scattered intensity versus wavelength from the bottom target(s) can be measured. We show that this far-field measurement can be used to determine the location of the target(s) with deep-subwavelength accuracy. In addition, discrimination of targets made from different materials can be achieved.
- 1 February 2019
- Published by AIP Publishing in APL Photonics
- Vol. 4 (2) , 026103
Open AccessNanoscale fingerprinting with hyperbolic metamaterials
We develop a new concept for optical sub-wavelength discrimination of nanostructures positioned on a hyperbolic metamaterial (HMM) surface using only far-field measurements of the spectral scattered intensity. The device demonstrating our approach consists of an HMM structure, one strong scatterer on top of the HMM at an a priori known location, and the target nanostructures to be sensed on the opposite interface of the HMM. Under far-field illumination from the top, highly localized beams are generated within the HMM by the top scatterer, with wavelength-dependent propagation directions. If the incident wavelength is varied, the localized beam will sweep across the target(s) and the far-field scattered intensity versus wavelength from the bottom target(s) can be measured. We show that this far-field measurement can be used to determine the location of the target(s) with deep-subwavelength accuracy. In addition, discrimination of targets made from different materials can be achieved.