Author Affiliations
1School of Information and Electronics, Beijing Institute of Technology, Beijing 100081, China2Advanced Research Institute of Multidisciplinary Sciences, Beijing Institute of Technology, Beijing 100081, China3Department of Automation, Tsinghua University, Beijing 100084, Chinashow less
Fig. 1. Flowchart of alternating projection algorithms. (a) GS algorithm; (b) input-output algorithm; (c) output-output algorithm; (d) hybrid input-output algorithm
Fig. 2. FPM reconstruction using AP algorithm
[25].
(a) FPM system structure; (b) FPM imaging model; (c) strength constraints imposed by AP algorithm in solution space; (d) AP solving FPM process
Fig. 3. Applications of FPM technique
[25]. (a) Quantitative phase imaging of blood smear; (b) phase images of live HeLa cell; (c) phase images of mitosis and apoptosis events of live HeLa cell captured by annular illumination FPM at a frame rate of 25 Hz; (d) reconstruction of three-dimensional refractive index of HeLa cells by FPM; (e) topographic map of a 3D surface via FP technology
Fig. 4. CDI techniques and reconstruction algorithms
[30]. (a) Plane-wave CDI; (b) Bragg CDI; (c) ptychographic CDI; (d) Fresnel CDI; (e) reflection CDI; (f) flowchart of alternating-projection-based CDI reconstruction algorithm
Fig. 5. Applications of CDI technique
[30]. (a) 3D mass density distribution of an unstained yeast spore cell; (b) 3D image of an unstained human chromosome; (c) reconstruction of an unstained herpesvirus virion; (d) quantitative 3D measurement of osteocyte; (e) representative diffraction pattern of a giant mimivirus particle; (f) 3D reconstruction of a mimivirus; (g) diffraction pattern of a nanocrystal; (h) electron density map of 2mFo-DFc
Fig. 6. Framework of DIP algorithm. (a) Principle of DIP; (b) common CNN structure; (c) UNet structure; (d) deep decoder structure
Fig. 7. Results of DIP algorithm and comparisons with other algorithms in each task. (a) Inpainting
[31]; (b) diffraction imaging
[51]; (c) phase unwarpping
[63] Fig. 8. Framework and applications of the PnP-GAP optimization. (a) Diagram of plug-and-play optimization framework based on GAP; (b) (c) comparison of large-scale snapshot compressive imaging and Fourier ptychographic microscopy between plug-and-play optimization and other methods, respectively
[6,90] Fig. 9. Fusion process of living glioblastoma observed by using plug-and-play optimization framework based on GAP
[6] Algorithm | Network | Input | Regularization | Application |
---|
Algorithm in Ref.[31] | UNet | Degraded image | / | Denoising & Inpainting & SR | Algorithm in Ref.[50] | CNN | PSF | / | FPM | Algorithm in Ref.[47] | CNN | Measurement | / | PR | Algorithm in Ref.[52] | UNet | Measurement | / | CT | Algorithm in Ref.[56] | UNet | Degraded image | TV | Denoising & Deblurring | Algorithm in Ref.[51] | UNet | Measurement | / | PR | Algorithm in Ref.[53] | DD | Noise | / | MRI | Algorithm in Ref.[61] | 2×CNN | Noise | / | MR correction | Algorithm in Ref.[60] | 3×DD | Noise | Dark channel prior | Retinal image enhancement | Algorithm in Ref.[57] | UNet | Initial reconstruction | TV | CT | Algorithm in Ref.[58] | UNet | Noise | Nonlocal | OCT | Algorithm in Ref.[59] | UNet | Reference image | / | MRI | Algorithm in Ref.[41] | 2×CNN+VAE | Haze image | HSV loss & Smooth loss on air-light map | Dehaze | Algorithm in Ref.[61] | 3×UNet | Noise | TV | Dynamic PET | Algorithm in Ref.[63] | UNet | Measurement | / | Phase unwarpping | Algorithm in Ref.[64] | DNN | To-be-updated image | / | SCI |
|
Table 1. DIP algorithms and their applications
Model | Framework | Fidelity term | Regularization term |
---|
| HQS | | | FISTA | | | ADMM | | | | GAP | | | AMP | | | RED | | |
|
Table 2. Plug-and-play optimization framework
Algorithm | Framework | Denoiser | Application |
---|
Algorithm in Ref.[75] | ADMM | BM3D | Tomography | Algorithm in Ref.[71] | FISTA(FASTA) | DnCNN | CDI & CDP | Algorithm in Ref.[74] | SGD & FISTA | TV & BM3D | FPM | Algorithm in Ref.[77] | ADMM | DnCNN | Tomography | Algorithm in Ref.[73] | ADMM & FISTA | DnCNN & MemNet & Residual Unet | MRI & CDP | Algorithm in Ref.[113] | GD | BM3D & FFDNET | CDP | Algorithm in Ref.[114] | RED | NLM & BM3D | Denoising & SR & Deblurring | Algorithm in Ref.[68] | HQS | SRResNet | SR | Algorithm in Ref.[90] | GAP | FFDNET | SCI | Algorithm in Ref.[115] | ADMM | DnCNN | CDI | Algorithm in Ref.[77] | GD | DnCNN | Tomography | Algorithm in Ref.[88] | ISTA | NLM | Inpainting & Deblurring | Algorithm in Ref.[116] | ADMM | BM3D | Hyperspectral PR | Algorithm in Ref.[6,117] | GAP | FFDNET | CDI & CDP & FPM & Pixel SR | Algorithm in Ref.[118] | GEC | Modified DnCNN | MRI |
|
Table 3. Plug-and-play optimization frameworks and their applications