Table 4 Quantitative comparisons between...

Table 4

Quantitative comparisons between MSAN and other methods in terms of the mean PSNR and SSIM on the W2S dataset [56]. The first and second best performances are denoted in red and blue, respectively. Methods marked with ‘*’ were trained with the training dataset [56].

Methods		PSNR(dB)/SSIM
		The number of raw images for averaging					Average value
		1	2	4	8	16	Average value
Traditional method	Noised	21.34 / 0.3612	23.78 / 0.4480	26.38 / 0.5473	29.10 / 0.6537	31.92 / 0.7555	26.50 / 0.5531
	NLM	23.68 / 0.7731	26.61 / 0.8204	29.69 / 0.8597	32.57 / 0.8899	35.29 / 0.9169	29.57 / 0.8520
	BM3D	24.18 / 0.8081	27.20 / 0.8508	30.33 / 0.8834	33.21 / 0.9075	36.00 / 0.9298	30.18 / 0.8760
	KSVD	23.83 / 0.7841	26.83 / 0.8307	29.99 / 0.8696	32.93 / 0.8989	35.80 / 0.9255	29.88 / 0.8617
	EPLL	24.09 / 0.7998	27.12 / 0.8454	30.25 / 0.8803	33.14 / 0.9059	35.94 / 0.9291	30.11 / 0.8721
	WNNM	24.01 / 0.8005	27.04 / 0.8450	30.20 / 0.8797	33.12 / 0.9053	35.97 / 0.9288	30.07 / 0.8719
	PURE-LET	24.09 / 0.8000	27.16 / 0.8434	30.29 / 0.8783	33.15 / 0.9026	35.89 / 0.9260	30.12 / 0.8701
Early DL	DnCNN*	33.51 / 0.9029	35.04 / 0.9189	37.23 / 0.9337	38.62 / 0.945	40.15 / 0.9540	36.91 / 0.9309
	IRCNN *	33.49 / 0.9094	34.94 / 0.9235	37.08 / 0.9356	38.68 / 0.9477	39.87 / 0.9557	36.81 / 0.9344
	MemNet*	31.45 / 0.8767	34.28 / 0.9121	35.76 / 0.9233	37.20 / 0.9319	39.17 / 0.9493	35.57 / 0.9187
SotA DL	Noise2Noise*	32.93 / 0.9055	35.19 / 0.9203	37.02 / 0.9313	38.41 / 0.9467	40.13 / 0.9556	36.74 / 0.9319
	MWCNN*	32.89 / 0.8946	34.69 / 0.9179	36.90 / 0.9337	38.31 / 0.9447	40.00 / 0.9547	36.56 / 0.9291
	RIDNet*	33.70 / 0.8707	34.91 / 0.8983	36.97 / 0.9254	38.53 / 0.9446	40.20 / 0.9589	36.86 / 0.9196
	DPDN*	33.51 / 0.8728	35.33 / 0.8915	37.07 / 0.9231	38.54 / 0.9456	39.81 / 0.9501	36.85 / 0.9166
	WF-UNet*	32.67 / 0.7366	34.31 / 0.8316	35.88 / 0.8507	37.04 / 0.8714	39.26 / 0.8895	35.83 / 0.8360
	MSAN* (Ours)	33.79 / 0.9102	35.57 / 0.9267	37.33 / 0.9378	38.81 / 0.9488	40.20 / 0.9538	37.14 / 0.9354

Methods		PSNR(dB)/SSIM
		The number of raw images for averaging					Average value
		1	2	4	8	16	Average value
Traditional method	Noised	21.34 / 0.3612	23.78 / 0.4480	26.38 / 0.5473	29.10 / 0.6537	31.92 / 0.7555	26.50 / 0.5531
	NLM	23.68 / 0.7731	26.61 / 0.8204	29.69 / 0.8597	32.57 / 0.8899	35.29 / 0.9169	29.57 / 0.8520
	BM3D	24.18 / 0.8081	27.20 / 0.8508	30.33 / 0.8834	33.21 / 0.9075	36.00 / 0.9298	30.18 / 0.8760
	KSVD	23.83 / 0.7841	26.83 / 0.8307	29.99 / 0.8696	32.93 / 0.8989	35.80 / 0.9255	29.88 / 0.8617
	EPLL	24.09 / 0.7998	27.12 / 0.8454	30.25 / 0.8803	33.14 / 0.9059	35.94 / 0.9291	30.11 / 0.8721
	WNNM	24.01 / 0.8005	27.04 / 0.8450	30.20 / 0.8797	33.12 / 0.9053	35.97 / 0.9288	30.07 / 0.8719
	PURE-LET	24.09 / 0.8000	27.16 / 0.8434	30.29 / 0.8783	33.15 / 0.9026	35.89 / 0.9260	30.12 / 0.8701
Early DL	DnCNN*	33.51 / 0.9029	35.04 / 0.9189	37.23 / 0.9337	38.62 / 0.945	40.15 / 0.9540	36.91 / 0.9309
	IRCNN *	33.49 / 0.9094	34.94 / 0.9235	37.08 / 0.9356	38.68 / 0.9477	39.87 / 0.9557	36.81 / 0.9344
	MemNet*	31.45 / 0.8767	34.28 / 0.9121	35.76 / 0.9233	37.20 / 0.9319	39.17 / 0.9493	35.57 / 0.9187
SotA DL	Noise2Noise*	32.93 / 0.9055	35.19 / 0.9203	37.02 / 0.9313	38.41 / 0.9467	40.13 / 0.9556	36.74 / 0.9319
	MWCNN*	32.89 / 0.8946	34.69 / 0.9179	36.90 / 0.9337	38.31 / 0.9447	40.00 / 0.9547	36.56 / 0.9291
	RIDNet*	33.70 / 0.8707	34.91 / 0.8983	36.97 / 0.9254	38.53 / 0.9446	40.20 / 0.9589	36.86 / 0.9196
	DPDN*	33.51 / 0.8728	35.33 / 0.8915	37.07 / 0.9231	38.54 / 0.9456	39.81 / 0.9501	36.85 / 0.9166
	WF-UNet*	32.67 / 0.7366	34.31 / 0.8316	35.88 / 0.8507	37.04 / 0.8714	39.26 / 0.8895	35.83 / 0.8360
	MSAN* (Ours)	33.79 / 0.9102	35.57 / 0.9267	37.33 / 0.9378	38.81 / 0.9488	40.20 / 0.9538	37.14 / 0.9354

[ViewLarge]