De-Scattering Algorithm for Underwater Mueller Matrix Images Based on Residual UNet

Xiaohuan Li; Xia Wang; Conghe Wang; Xin Zhang

doi:10.3788/AOS202242.2410001

Acta Optica Sinica, Volume. 42, Issue 24, 2410001(2022)

De-Scattering Algorithm for Underwater Mueller Matrix Images Based on Residual UNet

Xiaohuan Li, Xia Wang^*, Conghe Wang, and Xin Zhang

Key Laboratory of Photoelectronic Imaging Technology and System, Ministry of Education, School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(16)

Fig. 1. Model structure of designed Mu-UNet network

Download full size

Fig. 2. Schematic diagram of residual module Res Block in network. (a) Structure of residual module; (b) improved residual block structure with bottleneck layer

Download full size

Fig. 3. Experimental setup of underwater polarization imaging

Download full size

Fig. 4. Mueller matrix images of school badge target

Download full size

Fig. 5. Filtering results of different components by Canny arithmetic. (a) Origin image; (b) $σ = 1.0000$ ; (c) $σ = 0.0010$ ; (d) $σ = 0.0001$

Download full size

Fig. 6. Edge information and contour map of school badge image extracted by Canny operator and adaptive segmentation function. (a) Amount of edge information extracted by Canny operator; (b) area enclosed by contour obtained by adaptive threshold segmentation and contour extraction

Download full size

Fig. 7. 12 Mueller matrix images selected as network input

Download full size

Fig. 8. Original image of school badge target

Download full size

Fig. 9. Restoration results of underwater image under different turbidities. (a) Original underwater image with turbidity; (b) He's method; (c) Liang's method; (d) UNet network ; (e) Mu-UNet network; (f) ground truth

Download full size

Fig. 10. Statistical curves of pixel intensity values of line 128 at low concentration images in Fig. 9

Download full size

Fig. 11. Underwater restoration results of different material targets. (a) Original underwater image with turbidity; (b) He's method;(c) Liang's method; (d) UNet network ; (e) Mu-UNet network; (f) ground truth

Download full size

Fig. 12. Restoration results of untrained targets

Download full size

Table 1. LFF values of 16 components of Mueller matrix images of school badge

View table

Table 1. LFF values of 16 components of Mueller matrix images of school badge

Matrix element	$e_{c a}$	$a_{c t}$	$L F F$
$m_{11}$	3987	43.5	91.66
$m_{12}$	2235	63	35.47
$m_{13}$	1576	25	63.04
$m_{14}$	3838	204.5	18.77
$m_{21}$	6308	317.5	19.87
$m_{22}$	3945	94	41.97
$m_{23}$	3782	46	82.22
$m_{24}$	3567	295	12.09
$m_{31}$	2952	92.5	31.91
$m_{32}$	4518	337	13.41
$m_{33}$	3923	73.5	53.37
$m_{34}$	3828	509	7.52
$m_{41}$	657	43	15.28
$m_{42}$	3870	483.5	7.82
$m_{43}$	3051	290	10.52
$m_{44}$	1186	111	10.69

Table 2. Comparison of reconstruction results of proposed algorithm and other algorithms

View table

Table 2. Comparison of reconstruction results of proposed algorithm and other algorithms

Image No.	Index
	Original image		He's method		Liang's method		UNet		Mu-UNet
	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
1	12.37	0.21	10.74	0.07	9.08	0.19	12.42	0.26	16.32	0.35
2	13.22	0.26	8.15	0.15	16.57	0.51	15.82	0.46	20.22	0.54
3	9.60	0.48	6.55	0.22	13.3	0.43	19.57	0.69	20.14	0.75
4	11.81	0.61	4.23	0.23	9.56	0.46	13.32	0.63	20.27	0.75
5	8.51	0.27	8.03	0.21	9.23	0.26	16.11	0.58	18.61	0.72
6	10.97	0.50	5.63	0.18	12.48	0.36	25.23	0.83	27.58	0.87

Table 3. Comparison of reconstruction results by different models
View table
Table 3. Comparison of reconstruction results by different models
Model No. Train/test ratio PSNR SSIM
1 6∶1 20.52 0.66
2 4∶1 18.79 0.57
3 2∶1 16.66 0.51

Table 4. Running time of single image processed by different algorithms
View table
Table 4. Running time of single image processed by different algorithms
Method Times cost /s Platform
He′s 1.170 Matlab（CPU）
Liang′s 7.190 Matlab（CPU）
UNet 0.157 Pytorch（GPU）
Mu-UNet 0.215 Pytorch（GPU）

Tools

Get Citation

Copy Citation Text

Xiaohuan Li, Xia Wang, Conghe Wang, Xin Zhang. De-Scattering Algorithm for Underwater Mueller Matrix Images Based on Residual UNet[J]. Acta Optica Sinica, 2022, 42(24): 2410001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites