Deep prior-based sparse representation model for diffraction imaging: A plug-and-play method

Shi, Baoshun; Lian, Qiusheng; Chang, Huibin

doi:10.1016/j.sigpro.2019.107350

Cited by 23 publications

(14 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This assumption has been proven effective in ref. [4]. By doing this, the image filtering and image updating steps can be decoupled in the iteration process.…”

Section: Problem Formulation and Solvermentioning

confidence: 99%

See 1 more Smart Citation

Supervised dual tight frame learning with deep thresholding network for phase retrieval

Shi

Lian

2022

IET Image Processing

Self Cite

View full text Add to dashboard Cite

Data-driven tight frames are popular for solving imaging inverse problems. However, the imaging quality is limited by the representation ability of single tight frame and thresholds tuned manually. In this work, a supervised dual tight frame learning framework fused with an elaborated deep thresholding network (DTN) is proposed, and the issue of low-quality reconstructions in previous phase retrieval (PR) algorithms is addressed. To effectively learn dual tight frames, a loss function is formed using the mean square error, tight constraint, dual constraint, and sparse constraint terms. Moreover, to determine the thresholds adaptively, the thresholds are extracted from the frame coefficients via DTN. By an end-to-end supervised learning manner, the dual tight frames and DTN are jointly trained from labels and their counterparts corrupted by Gaussian noise. Using the Gaussian denoiser constructed by dual tight frames, a regularisation model is firstly designed, and then exploited to formulate a PR optimisation problem. The image filtering and image updating steps are performed alternatively for solving this problem. Particularly, the image updating subproblem is tackled by an inertial epigraph solver. The simulation experiments show that the proposed PR algorithm can obtain higher-quality reconstructions compared with the benchmark ones.

show abstract

“…This assumption has been proven effective in ref. [4]. By doing this, the image filtering and image updating steps can be decoupled in the iteration process.…”

Section: Problem Formulation and Solvermentioning

confidence: 99%

“…Phase retrieval (PR), i.e. the recovery of the original image from non-linear measurements, arises in many optical imaging applications [1][2][3][4]. Since the non-linear nature of the sampling operator, the imaging inverse problems are non-linear and non-convex.…”

Section: Introductionmentioning

confidence: 99%

Supervised dual tight frame learning with deep thresholding network for phase retrieval

Shi

Lian

2022

IET Image Processing

Self Cite

View full text Add to dashboard Cite

show abstract

“…Previous related work [15] [17] used the forward model of Section II-A in the PnP algorithm with a standard Gaussian denoiser as the prior model. PnP has also been used with deep neural network denoisers serving as prior models [22] [23] [24]. However, limitations of this previous work include little ability to control regularization strength and the use of generically trained neural networks as opposed to domainspecific denoisers.…”

Section: B Mace Reconstruction Frameworkmentioning

confidence: 99%

Multi-Resolution Data Fusion for Super Resolution Imaging

Reid¹,

Drummy²,

Bouman³

et al. 2021

Preprint

View full text Add to dashboard Cite

Applications in materials and biological imaging are limited by the ability to collect high-resolution data over large areas in practical amounts of time. One possible solution to this problem is to collect low-resolution data and interpolate to produce a high-resolution image. However, state-of-the-art super-resolution algorithms are typically designed for natural images, require aligned pairing of high and low-resolution training data for optimal performance, and do not directly incorporate a model of the imaging sensor.In this paper, we present a Multi-Resolution Data Fusion (MDF) algorithm for accurate interpolation of low-resolution SEM and TEM data by factors of 4x and 8x. This MDF interpolation algorithm uses small quantities of unpaired highresolution data to learn an accurate prior model denoiser and balances this with a forward model agent based on a mismatched back-projector that maintains fidelity to measured data. Our method is based on Multi-Agent Consensus Equilibrium, a generalization of the Plug-and-Play method, and allows for interpolation at arbitrary resolutions without retraining. We present electron microscopy results at 4x and 8x interpolation factors that exhibit reduced artifacts relative to existing methods while maintaining fidelity to acquired data and accurately resolving sub-pixel-scale features.

show abstract

“…This led to the work on D-AMP by Metzler [13] [42], using block-matching and 3D filtering (BM3D) and DnCNN denoisers. The BM3D and DnCNN denoisers have also been used in recent work on image CS reconstruction using the plug-and-play method [43] [44] [45].…”

Section: Related Workmentioning

confidence: 99%

Adaptive Block Compressive Sensing: Toward a Real-Time and Low-Complexity Implementation

Zammit¹,

Wassell²

2020

IEEE Access

View full text Add to dashboard Cite

Adaptive block-based compressive sensing (ABCS) algorithms are studied in the context of the practical realisation of compressive sensing on resource-constrained image and video sensing platforms that use single-pixel cameras, multi-pixel cameras or focal plane processing sensors. In this paper, we introduce two novel ABCS algorithms that are suitable for compressively sensing images or intra-coded video frames. Both use deterministic 2D-DCT dictionaries when sensing the images instead of random dictionaries. The first uses a low number of compressive measurements to compute the block boundary variation (BBV) around each image block, from which it estimates the number of 2D-DCT transform coefficients to measure from each block. The second uses a low number of DCT domain (DD) measurements to estimate the total number of transform coefficients to capture from each block. The two algorithms permit reconstruction in real time, averaging 8 ms and 26 ms for 256 × 256 and 512 × 512 greyscale images, respectively, using a simple inverse 2D-DCT operation without requiring GPU acceleration. Furthermore, we show that an iterative compressive sensing reconstruction algorithm (IDA), inspired by the denoisingbased approximate message passing algorithm, can be used as a post-processing, quality enhancement technique. IDA trades off real-time operation to yield performance improvement over state-of-the-art GPUassisted algorithms of 1.31 dB and 0.0152 in terms of PSNR and SSIM, respectively. It also exceeds the PSNR performance of a state-of-the-art deep neural network by 0.4 dB and SSIM by 0.0126.

show abstract

Deep prior-based sparse representation model for diffraction imaging: A plug-and-play method

Cited by 23 publications

References 42 publications

Supervised dual tight frame learning with deep thresholding network for phase retrieval

Supervised dual tight frame learning with deep thresholding network for phase retrieval

Multi-Resolution Data Fusion for Super Resolution Imaging

Adaptive Block Compressive Sensing: Toward a Real-Time and Low-Complexity Implementation

Contact Info

Product

Resources

About