Image Restoration Based on Adaptive Dual-Domain Filtering

Table 4

PSNR comparison on 9 color images from the datasets LIVE1 [15] and Set14 [16] for Gaussian denoising. One can see that the proposed ADDF really outperforms the state-of-the-art methods BM3D [5], DDID [9], and DDF [27].

	Methods	Lena	Peppers	Girl	Kodak2	Kodak7	Kodak10	Kodak14	Kodak19	Kodak23	Avg.
	BM3D	32.27	31.20	31.33	32.44	33.66	33.82	30.03	31.78	34.79	32.37
	DDID	32.30	31.25	31.45	32.29	33.52	33.70	30.03	31.57	34.83	32.33
	DDF	32.45	31.41	31.58	32.49	33.74	33.89	30.18	31.72	34.97	32.49
	ADDF	32.45	31.40	31.58	32.47	33.77	33.90	30.18	31.71	34.99	32.49
	BM3D	30.11	29.27	29.16	30.30	30.60	30.92	27.48	29.36	32.11	29.92
	DDID	30.68	29.88	29.58	30.46	31.09	31.46	27.82	29.66	32.75	30.38
	DDF	30.87	30.05	29.70	30.66	31.34	31.66	27.97	29.83	32.93	30.56
	ADDF	30.90	30.05	29.72	30.67	31.42	31.74	27.98	29.84	33.01	30.59
	BM3D	29.54	28.58	28.25	29.51	29.52	30.05	26.35	28.71	31.32	29.09
	DDID	29.44	28.78	28.22	29.25	29.37	29.88	26.40	28.45	31.25	29.00
	DDF	29.66	28.97	28.38	29.48	29.64	30.11	26.56	28.68	31.48	29.22
	ADDF	29.75	29.01	28.44	29.54	29.78	30.25	26.60	28.72	31.60	29.30
	BM3D	28.63	27.66	27.29	28.62	28.35	28.97	25.40	27.86	30.20	28.11
	DDID	28.41	27.82	27.17	28.33	28.08	28.67	25.35	27.52	30.06	27.93
	DDF	28.66	28.02	27.35	28.59	28.36	28.92	25.51	27.80	30.33	28.17
	ADDF	28.80	28.13	27.45	28.70	28.54	29.10	25.57	27.90	30.54	28.30
	BM3D	27.88	26.87	26.54	27.87	27.43	28.12	24.65	27.17	29.26	27.31
	DDID	27.52	26.95	26.30	27.57	27.03	27.70	24.50	26.73	29.07	27.04
	DDF	27.79	27.16	26.52	27.85	27.31	27.97	24.66	27.07	29.37	27.30
	ADDF	27.97	27.31	26.64	28.01	27.54	28.20	24.79	27.21	29.60	27.47
	BM3D	27.11	26.06	25.85	27.06	26.44	27.25	23.89	26.38	28.32	26.48
	DDID	26.75	26.16	25.58	26.91	26.16	26.89	23.81	26.04	28.21	26.28
	DDF	27.04	26.38	25.84	27.22	26.43	27.18	23.97	26.41	28.53	26.56
	ADDF	27.26	26.58	25.97	27.42	26.69	27.44	24.10	26.61	28.82	26.77