Self-Supervised Denoising Method for Polarization Image Based on Mask Strategy (Invited)

Haofeng Hu; Tianci Li; Yuanyang Bu; Linghao Shen

doi:10.3788/LOP251422

Laser & Optoelectronics Progress, Volume. 62, Issue 19, 1906016(2025)

Self-Supervised Denoising Method for Polarization Image Based on Mask Strategy (Invited)

Haofeng Hu^1,2,3, Tianci Li^1,3, Yuanyang Bu⁴, and Linghao Shen^2、*

Author Affiliations

¹School of Future Technology, Tianjin University, Tianjin 300072, China

²School of Marine Science and Technology, Tianjin University, Tianjin 300072, China

³School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China

⁴School of Automation, Northwestern Polytechnical University, Xi'an 710072, Shaanxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Overall framework of polarization denoising algorithm

Download full size

Fig. 2. Comparison of polarization denoising results by different denoising algorithms

Download full size

Fig. 3. Comparison of denoising results before and after adding AoP loss. (a) Recovery images; (b) detail enlargement images

Download full size

Fig. 4. Effects of different mask ratios on polarization denoising results

Download full size

Fig. 5. Denoising performances on the green channel of color-polarization dataset by proposed algorithm when adding G&P Noise (σ = λ = 20.0)

Download full size

Fig. 6. Denoising performances on color-polarization dataset by proposed algorithm when adding G&P Noise (σ = λ = 20.0)

Download full size

Table 1. PSNR of different denoising algorithms applied to polarization denoising

View table

Table 1. PSNR of different denoising algorithms applied to polarization denoising

Test noise	Image	Traditional	Supervised		Unsupervised				Noise
Test noise	Image	BM3D	DnCNN	Restormer	Noise2Void	AP-BSN	SDAP	Proposed	Noise
Real moise （σ = 14.7）	S₀	25.68	25.41	27.28	25.43	25.55	25.84	26.44	23.84
	DoP	22.72	26.34	22.73	21.99	22.84	25.30	27.31	8.21
	AoP	9.95	10.94	8.59	8.69	9.17	10.24	13.30	7.90
Real noise （σ = 16.1）	S₀	32.85	33.82	30.94	31.87	33.23	32.50	34.07	25.54
	DoP	23.61	26.23	23.90	22.91	23.90	25.95	27.92	9.25
	AoP	10.28	10.22	8.68	8.65	9.26	10.53	13.63	7.94
AWGN （σ = 30.0）	S₀	32.01	33.27	27.31	26.56	30.27	30.59	35.76	19.22
	DoP	21.17	26.21	13.92	19.43	21.42	22.41	27.14	4.27
	AoP	8.74	10.12	8.84	8.39	8.44	8.66	13.35	7.78
Poisson noise （λ = 25.0）	S₀	33.21	34.37	29.97	29.83	31.18	31.66	39.00	20.38
	DoP	20.10	26.16	22.14	20.21	21.81	22.38	27.02	5.03
	AoP	8.78	10.22	8.39	8.28	8.47	8.63	13.44	7.83
G&P Noise （σ = λ = 20.0）	S₀	30.37	33.98	26.58	27.24	29.48	29.53	36.50	17.83
	DoP	18.38	25.93	21.92	21.34	20.45	20.52	27.25	3.36
	AoP	8.28	9.72	8.25	8.22	8.21	8.23	13.32	7.74
S&P Noise （r = 0.12）	S₀	20.96	14.82	18.11	20.79	21.15	19.82	23.98	10.90
	DoP	11.20	6.17	18.91	14.04	15.53	11.53	25.78	3.23
	AoP	7.80	7.61	7.93	8.05	7.80	7.62	12.89	9.71
S&P Noise （r = 0.05）	S₀	25.80	19.69	23.23	26.03	23.88	24.60	28.48	13.47
	DoP	14.78	9.46	14.64	18.27	14.34	13.65	26.49	5.39
	AoP	7.95	8.09	7.86	8.37	7.90	8.06	13.16	11.46

Table 2. SSIM of different denoising algorithms applied to polarization denoising

View table

Table 2. SSIM of different denoising algorithms applied to polarization denoising

Test noise	Image	Traditional	Supervised		Unsupervised				Noise
Test noise	Image	BM3D	DnCNN	Restormer	Noise2Void	AP-BSN	SDAP	Proposed	Noise
Real moise （σ = 14.7）	S₀	0.883	0.857	0.858	0.860	0.884	0.884	0.927	0.416
	DoP	0.550	0.616	0.308	0.372	0.440	0.545	0.682	0.013
	AoP	0.151	0.213	0.056	0.052	0.063	0.162	0.353	0.028
Real noise （σ = 16.1）	S₀	0.920	0.929	0.859	0.858	0.917	0.914	0.958	0.396
	DoP	0.582	0.585	0.350	0.387	0.474	0.579	0.703	0.019
	AoP	0.190	0.191	0.062	0.051	0.069	0.183	0.407	0.031
AWGN （σ = 30.0）	S₀	0.806	0.857	0.447	0.73	0.790	0.819	0.906	0.112
	DoP	0.335	0.582	0.194	0.264	0.241	0.335	0.662	0.002
	AoP	0.046	0.158	0.065	0.033	0.028	0.039	0.367	0.016
Poisson noise （λ = 25.0）	S₀	0.842	0.885	0.775	0.778	0.837	0.846	0.949	0.197
	DoP	0.361	0.585	0.251	0.239	0.249	0.323	0.675	0.002
	AoP	0.064	0.161	0.046	0.025	0.034	0.044	0.382	0.020
G&P Noise （σ = λ = 20.0）	S₀	0.737	0.869	0.711	0.763	0.760	0.768	0.918	0.104
	DoP	0.213	0.568	0.243	0.290	0.174	.204	0.662	0.001
	AoP	0.037	0.156	0.035	0.032	0.024	0.029	0.364	0.015
S&P Noise （r = 0.12）	S₀	0.406	0.057	0.494	0.342	0.414	0.229	0.808	0.021
	DoP	0.055	0.002	0.196	0.032	0.072	0.016	0.607	0.0001
	AoP	0.020	0.011	0.013	0.011	0.012	0.011	0.295	0.159
S&P Noise （r = 0.05）	S₀	0.638	0.366	0.449	0.670	0.477	0.488	0.871	0.051
	DoP	0.164	0.107	0.065	0.157	0.089	0.073	0.640	0.001
	AoP	0.036	0.024	0.014	0.024	0.014	0.017	0.346	0.247

Table 3. PSNR and SSIM before and after adding AoP loss under different noise levels

View table

Table 3. PSNR and SSIM before and after adding AoP loss under different noise levels

Test noise	PSNR /dB		SSIM
Test noise	With AoP loss	Without AoP loss	With AoP loss	Without AoP loss
Real moise （σ = 14.7）	13.30	12.56	0.353	0.367
Real noise （σ = 16.1）	13.63	12.49	0.407	0.391
AWGN （σ = 30.0）	13.35	12.52	0.367	0.357
Poisson noise （λ = 25.0）	13.44	12.83	0.382	0.380
G&P Noise （σ = λ = 20.0）	13.32	12.51	0.364	0.354

Table 4. Results of ablation experiments for each component

View table

Table 4. Results of ablation experiments for each component

without Pretrain	without Stack	Last	EMA after Iteration	PSNR /dB			SSIM
without Pretrain	without Stack	Last	EMA after Iteration	S₀ image	DoP image	AoP image	S₀ image	DoP image	AoP image
√	×	×	×	33.85	27.55	12.66	0.956	0.694	0.392
×	√	×	×	33.26	24.83	10.84	0.927	0.550	0.241
×	×	√	×	33.35	27.70	13.39	0.948	0.691	0.384
×	×	×	√	33.87	27.95	13.59	0.956	0.706	0.405
×	×	×	×	34.07	27.92	13.63	0.958	0.703	0.407

Table 5. PSNR of green-channel denoising on color-polarization dataset by proposed algorithm

View table

Table 5. PSNR of green-channel denoising on color-polarization dataset by proposed algorithm

Test noise	Denoise result			Noise
Test noise	S₀ image	DoP image	AoP image	S₀ image	DoP image	AoP image
AWGN （σ = 30.0）	36.05	30.51	10.56	19.20	3.827	8.464
Poisson noise （λ = 25.0）	38.13	30.72	10.80	20.32	4.762	8.488
G&P Noise （σ = λ = 20.0）	36.12	30.53	10.38	17.82	3.223	8.441

Table 6. SSIM of green-channel denoising on color-polarization dataset by proposed algorithm

View table

Table 6. SSIM of green-channel denoising on color-polarization dataset by proposed algorithm

Test noise	Denoise result			Noise
Test noise	S₀ image	DoP image	AoP image	S₀ image	DoP image	AoP image
AWGN （σ = 30.0）	0.922	0.680	0.196	0.127	0.001	0.002
Poisson noise （λ = 25.0）	0.937	0.680	0.187	0.183	0.001	0.018
G&P Noise （σ = λ = 20.0）	0.921	0.675	0.185	0.110	0.001	0.015

Table 7. PSNR of denoising on color-polarization dataset by proposed algorithm

View table

Table 7. PSNR of denoising on color-polarization dataset by proposed algorithm

Test noise	Denoise result			Noise
Test noise	S₀ image	DoP image	AoP image	S₀ image	DoP image	AoP image
AWGN （σ = 30.0）	36.38	10.81	31.17	19.12	8.588	4.159
Poisson noise （λ = 25.0）	38.16	11.00	31.74	29.92	8.602	5.004
G&P Noise （σ = λ = 20.0）	36.33	10.59	31.02	17.52	8.561	3.438

Table 8. SSIM of denoising on color-polarization dataset by proposed algorithm

View table

Table 8. SSIM of denoising on color-polarization dataset by proposed algorithm

Test noise	Denoise result			Noise
Test noise	S₀ image	DoP image	AoP image	S₀ image	DoP image	AoP image
AWGN （σ = 30.0）	0.934	0.225	0.730	0.132	0.017	0.002
Poisson noise （λ = 25.0）	0.946	0.236	0.742	0.179	0.018	0.002
G&P Noise （σ = λ = 20.0）	0.932	0.204	0.721	0.110	0.015	0.001

Tools

Get Citation

Copy Citation Text

Haofeng Hu, Tianci Li, Yuanyang Bu, Linghao Shen. Self-Supervised Denoising Method for Polarization Image Based on Mask Strategy (Invited)[J]. Laser & Optoelectronics Progress, 2025, 62(19): 1906016

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Fiber Optics and Optical Communications

Received: Jun. 11, 2025

Accepted: Jul. 21, 2025

Published Online: Sep. 28, 2025

The Author Email: Linghao Shen (shenlinghao@tju.edu.cn)

DOI:10.3788/LOP251422

CSTR:32186.14.LOP251422

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology