Chengshuai Yang1, Yunhua Yao1、6、*, Chengzhi Jin1, Dalong Qi1, Fengyan Cao1, Yilin He1, Jiali Yao1, Pengpeng Ding1, Liang Gao2, Tianqing Jia1, Jinyang Liang3, Zhenrong Sun1, and Shian Zhang1、4、5、7、*
Author Affiliations
1State Key Laboratory of Precision Spectroscopy, School of Physics and Electronic Science, East China Normal University, Shanghai 200062, China2Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois 61801, USA3Institut National de la Recherche Scientifique, Centre Énergie Matériaux Télécommunications, Laboratory of Applied Computational Imaging, Varennes, Québec J3X1S2, Canada4Collaborative Innovation Center of Extreme Optics, Shanxi University, Taiyuan 030006, China5Collaborative Innovation Center of Light Manipulations and Applications, Shandong Normal University, Jinan 250358, China6e-mail: yhyao@lps.ecnu.edu.cn7e-mail: sazhang@phy.ecnu.edu.cnshow less
DOI: 10.1364/PRJ.410018
Cite this Article
Set citation alerts
Chengshuai Yang, Yunhua Yao, Chengzhi Jin, Dalong Qi, Fengyan Cao, Yilin He, Jiali Yao, Pengpeng Ding, Liang Gao, Tianqing Jia, Jinyang Liang, Zhenrong Sun, Shian Zhang. High-fidelity image reconstruction for compressed ultrafast photography via an augmented-Lagrangian and deep-learning hybrid algorithm[J]. Photonics Research, 2021, 9(2): B30
Copy Citation Text
show less
Fig. 1. Data flow chart of the AL+DL algorithm. (a) Solver Sp(·) in a sparse domain; (b) general framework by connecting each iteration in a sequence order. Here, each Si(·) is calculated in parallel to producing Wi, and GD algorithm is employed to calculate I.
Fig. 2. (a) U-net architecture in the AL+DL algorithm; (b) self-attention model.
Fig. 3. Reconstructed results of (a) boatman, (b) ocean animal, and (c) finger by the AL+DL (second row), AL (third row), and TwIST (fourth row) algorithms, together with the ground truth (first row) for comparison. The last column is the enlarged image in the corresponding red squares.
Fig. 4. System configuration of CUP. DMD, digital micromirror device; CMOS, complementary metal–oxide-semiconductor.
Fig. 5. Measuring temporal evolution of a spatially modulated picosecond laser spot. (a) Experimental design. (b)–(d) Reconstructed results by the AL+DL, AL, and TwIST algorithms, respectively. (e) Measured static image by external CCD. (f)–(h) Extracted images from (b)–(d), respectively, at the time of 14 ps; curves on the right are the integration results of the corresponding images along the horizontal direction.
Fig. 6. Measuring wavefront movement by obliquely illuminating a collimated femtosecond laser pulse on a transverse fan pattern. (a) Experimental design. (b)–(d) Reconstructed results by the AL+DL, AL, and TwIST algorithms, respectively. (e) Measured static image by external CCD. (f)–(h) Integrated images from (b)–(d), respectively. (i)–(l) Results of Fourier transform from (e)–(h), respectively.
Scene | | AL | TwIST |
---|
PSNR | SSIM | PSNR | SSIM | PSNR | SSIM |
---|
Boatman | 28.50 | 0.836 | 24.15 | 0.700 | 22.47 | 0.589 | Ocean animal | 30.47 | 0.916 | 25.00 | 0.802 | 24.72 | 0.781 | Finger | 42.00 | 0.983 | 32.22 | 0.932 | 28.56 | 0.894 |
|
Table 1. Average PSNR (in dB) and SSIM by Different Image Reconstruction Algorithms in Different Dynamic Scenes