Speckle denoising based on deep learning via a conditional generative adversarial network in digital holographic interferometry

Fang, Qiang; Xia, Haiting; Song, Qing; zhang, Meijuan; Guo, Rongxin; Montrésor, Silvio; Picart, Pascal

doi:10.1364/oe.459213

Cited by 19 publications

(8 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…: 40 pairs (15,360 patches) l 2 -norm Tahon et al 214 , 215 Noisy sine and cosine Noise-free sine and cosine DnCNN Sim. : 25 pairs and 128 pairs l 2 -norm Fang et al 216 Noisy real, imaginary Noise-free real, imaginary U-Net Sim. : 4000 pairs GAN loss Murdaca et al 217 Noisy real, imaginary, and amplitude Noise-free real, imaginary, and amplitude U-Net Sim.…”

Section: Dl-post-processing For Phase Recoverymentioning

confidence: 99%

See 1 more Smart Citation

On the use of deep learning for phase recovery

Wang,

Song,

Wang

et al. 2024

Light Sci Appl

View full text Add to dashboard Cite

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

show abstract

Section: Dl-post-processing For Phase Recoverymentioning

confidence: 99%

“…To go further, they released a more comprehensive dataset for conditions of severe speckle noise 215 . Fang et al 216 applied GAN to do speckle noise reduction for phase. Murdaca et al 217 applied this deep-learning-based phase noise reduction to interferometric synthetic aperture radar (InSAR) 218 .…”

Section: Dl-post-processing For Phase Recoverymentioning

confidence: 99%

On the use of deep learning for phase recovery

Wang,

Song,

Wang

et al. 2024

Light Sci Appl

View full text Add to dashboard Cite

show abstract

“…In the last decade, with the application of deep learning in optical metrology [31], such as image denoising [32,33], fringe pattern enhancement [34], wrapped phase retrieval [35], phase unwrapping [36], and stereo matching [37], many researchers have applied deep learning to FPP in order to obtain accurate wrapped phases with fewer patterns and further achieve phase unwrapping. Qian et al [38] proposed a single-shot absolute 3D shape measurement with deeplearning-based color FPP.…”

Section: Introductionmentioning

confidence: 99%

Three 1-bit speckle-embedded pulse-width modulation patterns for robust absolute 3D measurement

Zheng,

Li,

Zuo

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

In 3D shape measurement techniques using structured light, 1-bit pulse-width modulation (PWM) patterns and 1-bit speckle patterns can be projected at high speed. However, when combining PWM and speckle patterns to integrate their advantages, the decoupling problem is insurmountable. In this work, a novel 1-bit speckle-embedded pulse-width-modulation (SP-PWM) patterns was proposed to achieve absolute 3D shape measurement using only three binary patterns. Our method consists of three main steps: First, a sinusoidal pattern reconstruction network (SPRNet) was proposed to eliminate the high-order harmonics and speckle patterns in the SP-PWM patterns and obtain high-quality sinusoidal patterns. Second, a multi-temporal spatial correlation matching algorithm (MTSCMA) was proposed to obtain a coarse disparity map from the three SP-PWM patterns. Third, the high-accuracy wrapped phase map is used as an additional constraint for refining the coarse disparity map to obtain the final high-accuracy disparity map for absolute 3D measurement without phase unwrapping. Our method combines the advantages of fringe projection profilometry techniques for high-precision wrapped phase retrieval and speckle correlation matching algorithms for robust and unambiguous disparity map calculation. The experimental results demonstrated that our method could realize high-precision absolute 3D shape measurement with an accuracy of 0.057mm using only three 1-bit SP-PWM patterns. Furthermore, different simulation noises were used to demonstrate the robustness of the proposed method.

show abstract

“…However, traditional denoising methods that target Gaussian characteristics are less effective in speckle noise, which has non-Gaussian and non-statistical characteristics [7]. The implementation of deep learning methods presents a viable resolution to this issue [8,9]. In our previous work [8], a speckle denoising approach via Conditional Generative Adversarial Networks (cGANs) for speckle denoising is proposed, which is trained by using a paired dataset.…”

Section: Introductionmentioning

confidence: 99%

“…The implementation of deep learning methods presents a viable resolution to this issue [8,9]. In our previous work [8], a speckle denoising approach via Conditional Generative Adversarial Networks (cGANs) for speckle denoising is proposed, which is trained by using a paired dataset. However, directly obtaining speckle-free holograms experimentally as a reference for model training is unattainable, it makes building the dataset a challenging task.…”

Section: Introductionmentioning

confidence: 99%

Speckle denoising of digital holographic interferometry via integrated CycleGAN

Yu,

Fang,

Song

et al. 2023

Holography, Diffractive Optics, and Applications XIII

Self Cite

View full text Add to dashboard Cite

Digital Holographic Interferometry (DHI) provides a non-contact measurement at the wavelength level of light. Due to the application of the interferometric principle, non-Gaussian speckle noise introduced during the measurement process is unavoidable and is difficult to eliminate. Thus, denoising is critical and affects measurement accuracy. A method for speckle denoising via self-supervised deep learning, based on a Cycle-Generative Adversarial Network (CycleGAN), is proposed in this paper. The method employs unpaired datasets and integrates a 4-f optical speckle noise simulation module to reduce training costs while improving training accuracy. The proposed method was tested on both simulated and experimental data, with results showing a 4.6% performance improvement in PSNR over competitor algorithms. The proposed method has great potential and advantages in DHI studies with huge datasets.

show abstract

Speckle denoising based on deep learning via a conditional generative adversarial network in digital holographic interferometry

Cited by 19 publications

References 37 publications

On the use of deep learning for phase recovery

On the use of deep learning for phase recovery

Three 1-bit speckle-embedded pulse-width modulation patterns for robust absolute 3D measurement

Speckle denoising of digital holographic interferometry via integrated CycleGAN

Contact Info

Product

Resources

About