Pixelwise Estimation of Signal-Dependent Image Noise Using Deep Residual Learning

Table 6

Applying noise estimation methods to deep learning models to remove nonhomogeneous noise (dB).

Dataset	Noise level	Chen [7] + DRDD	DRNE + DRDD	Chen + FFDNet	DRNE + FFDNet
Kodak		39.91	39.90	39.84	39.84
		34.97	34.98	34.97	34.98
		33.04	33.07	33.10	33.12
		36.29	36.31	36.38	36.40
	Noise model (1)	33.30	33.63	33.24	33.68
McMaster		39.62	39.63	39.30	39.32
		35.04	35.05	34.93	34.95
		32.30	32.39	32.38	32.45
		34.66	34.72	34.61	34.67
	Noise model (1)	33.21	33.18	33.14	33.14
BSD500		39.47	39.46	39.44	39.43
		34.28	34.29	34.25	34.26
		32.34	32.38	32.35	32.40
		35.66	35.67	35.64	35.66
	Noise model (1)	33.01	33.20	32.92	33.17

Text in bold denotes the best method. Text in italics denotes significant performance gains of the DRNE against Chen’s method.