Enhancement Method of Image Quality for Lensless Imaging Based on Physics-Driven Unsupervised Learning

Jiale Zuo; Mengmeng Zhang; Ju Tang; Jiawei Zhang; Zhenbo Ren; Jianglei Di; Jianlin Zhao

doi:10.3788/AOS240742

Acta Optica Sinica, Volume. 44, Issue 16, 1611001(2024)

Enhancement Method of Image Quality for Lensless Imaging Based on Physics-Driven Unsupervised Learning

Jiale Zuo¹, Mengmeng Zhang¹, Ju Tang¹, Jiawei Zhang¹, Zhenbo Ren^1,2、*, Jianglei Di^3、**, and Jianlin Zhao¹

¹Key Laboratory of Light Field Manipulation and Information Acquisition, Ministry of Industry and Information Technology, Shaanxi Key Laboratory of Optical Information Technology, School of Physical Science and Technology, Northwestern Polytechnical University, Xi’an 710129, Shaanxi , China

²Research & Development Institute of Northwestern Polytechnical University in Shenzhen, Shenzhen 518063, Guangdong , China

³School of Information Engineering, Institute of Advanced Photonics Technology, Guangdong Provincial Key Laboratory of Information Photonics Technology, Guangdong University of Technology, Guangzhou 510006, Guangdong , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(17)

Fig. 1. Optical setup of lensless imaging system

Download full size

Fig. 2. Structure of DPNNA method

Download full size

Fig. 3. Variation curve of coefficient c

Download full size

Fig. 4. Structure of UNet 3+

Download full size

$Simulated results of phase object. (a) Object phase distribution; (b) in-line hologram pattern with diffraction distance of 12 mm; (c) in-line hologram pattern with diffraction distance of 22.4 mm; (d) result of G-S algorithm; (e) result of PhysenNet method; (f) result of DPNNA method$

Fig. 5. Simulated results of phase object. (a) Object phase distribution; (b) in-line hologram pattern with diffraction distance of 12 mm; (c) in-line hologram pattern with diffraction distance of 22.4 mm; (d) result of G-S algorithm; (e) result of PhysenNet method; (f) result of DPNNA method

Download full size

$Simulated results of phase object with irregular change in phase value. (a) Phase distribution; (b) in-line hologram patterns with diffraction distance of 12 mm; (c) in-line hologram patterns with diffraction distance of 22 mm; (d) results of G-S algorithm; (e) results of PhysenNet method; (f) results of DPNNA method$

Fig. 6. Simulated results of phase object with irregular change in phase value. (a) Phase distribution; (b) in-line hologram patterns with diffraction distance of 12 $m m$ ; (c) in-line hologram patterns with diffraction distance of 22 $m m$ ; (d) results of G-S algorithm; (e) results of PhysenNet method; (f) results of DPNNA method

Download full size

Fig. 7. Simulated results of amplitude object. (a) Object amplitude distribution; (b) result of DPNNA method; (c) local magnification of orange area; (d) local magnification of blue area; (e) grayscale changes of dashed lines in ground truth and DPNNA result

Download full size

$Experimental results of phase object. (a) In-line hologram patterns with diffraction distance of 3.00 mm; (b) in-line hologram patterns with diffraction distance of 3.01 mm; (c) results of DH algorithm; (d) results of G-S algorithm; (e) results of PhysenNet method; (f) results of DPNNA method$

Fig. 8. Experimental results of phase object. (a) In-line hologram patterns with diffraction distance of 3.00 $m m$ ; (b) in-line hologram patterns with diffraction distance of 3.01 $m m$ ; (c) results of DH algorithm; (d) results of G-S algorithm; (e) results of PhysenNet method; (f) results of DPNNA method

Download full size

Fig. 9. Grayscale changes in experimental results of DPNNA method

Download full size

Fig. 10. Experimental results of amplitude object. (a) Local magnification result of DH method; (b) local magnification result of AS method; (c) local magnification result of DPNNA method

Download full size

$Simulated results with diffraction interval of 100 mm. (a) Ground-truth; (b) in-line hologram pattern with diffraction distance of 12 mm; (c) in-line hologram pattern with diffraction distance of 112 mm; (d) result of DPNNA method$

Fig. 11. Simulated results with diffraction interval of 100 mm. (a) Ground-truth; (b) in-line hologram pattern with diffraction distance of 12 mm; (c) in-line hologram pattern with diffraction distance of 112 mm; (d) result of DPNNA method

Download full size

Fig. 12. Experimental results of 1951 USAF resolution test chart. (a) Result of DH algorithm; (b) result of AS algorithm; (c) result of DPNNA method

Download full size

$Comparison of phase recovery results under different noise levels. (a) In-line hologram pattern with SNR of 23.7 dB at diffraction distance of 12 mm; (b) in-line hologram pattern with SNR of 23.7 dB at diffraction distance of 22 mm; (c) result of DPNNA method; (d) absolute error between DPNNA and ground-truth; (e) local magnification of blue area; (f) local magnification of orange area$

Fig. 13. Comparison of phase recovery results under different noise levels. (a) In-line hologram pattern with SNR of 23.7 dB at diffraction distance of 12 mm; (b) in-line hologram pattern with SNR of 23.7 dB at diffraction distance of 22 mm; (c) result of DPNNA method; (d) absolute error between DPNNA and ground-truth; (e) local magnification of blue area; (f) local magnification of orange area

Download full size

Table 1. Quantitative error comparison of phase retrieval results
View table
Table 1. Quantitative error comparison of phase retrieval results
Evaluation indicator G-S PhysenNet DPNNA
PSNR /dB 23.47 21.03 46.63
SSIM 0.4330 0.8210 0.9998

Table 2. Quantitative error results
View table
Table 2. Quantitative error results
Object Evaluation indicator G-S PhysenNet DPNNA
Pattern 1 SSIM 0.482 0.682 0.940
PSNR /dB 7.60 10.19 27.81
Pattern 2 SSIM 0.830 0.813 0.756
PSNR /dB 18.58 22.45 14.83
Pattern 3 SSIM 0.696 0.817 0.892
PSNR /dB 14.25 17.97 16.21

Table 3. Quantitative error results under different diffraction intervals
View table
Table 3. Quantitative error results under different diffraction intervals
Evaluation indicator Diffraction interval
10^-4 mm 10^-3 mm 10^-2 mm 10^-1 mm 1 mm 10 mm 10² mm
PSNR /dB 45.67 41.41 41.17 51.86 40.73 48.76 40.59
SSIM 0.9997 0.9996 0.9996 0.9999 0.9996 0.9998 0.9996

Table 4. Comparison of quantitative results of DPNNA method under different noise levels
View table
Table 4. Comparison of quantitative results of DPNNA method under different noise levels
Evaluation indicator Value
SNR /dB 39.2 33.4 30.0 27.8 26.8 23.7
PSNR /dB 43.05 40.78 42.48 39.09 32.32 27.87
SSIM 0.9998 0.9996 0.9995 0.9988 0.9950 0.9880

Tools

Get Citation

Copy Citation Text

Jiale Zuo, Mengmeng Zhang, Ju Tang, Jiawei Zhang, Zhenbo Ren, Jianglei Di, Jianlin Zhao. Enhancement Method of Image Quality for Lensless Imaging Based on Physics-Driven Unsupervised Learning[J]. Acta Optica Sinica, 2024, 44(16): 1611001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites