An Efficient Universal Noise Removal Algorithm Combining Spatial Gradient and Impulse Statistic

Table 3

Comparative restoration results in PSNR (dB) and MAE (the second row) for random-valued impulse noise.

Method	Lena image				Boats image
	𝑝=20%	𝑝=30%	𝑝=40%	𝑝=50%	𝑝=20%	𝑝=30%	𝑝=40%	𝑝=50%
Noisy image	16.25	14.48	13.19	12.27	15.99	14.25	12.94	11.97
	6.923	10.41	13.96	17.26	8.332	12.37	16.75	20.88
3 × 3 median filter	31.54	28.17	24.68	21.60	29.22	26.55	23.46	20.48
	1.770	2.450	3.692	5.633	2.216	2.846	4.003	5.928
5 × 5 median filter	30.14	29.11	27.77	25.53	27.31	26.64	25.56	23.61
	2.213	2.606	3.214	4.300	3.085	3.349	3.868	5.068
SDROM	35.39	33.44	31.48	29.51	32.40	30.54	28.88	27.43
	0.620	0.972	1.362	1.830	1.195	1.586	2.279	2.864
ROAD-TRIF	34.66	33.05	31.19	29.43	31.85	30.25	28.65	27.41
	1.004	1.171	1.505	1.919	1.477	1.711	2.381	2.869
ROLD-EPR	34.84	33.15	31.28	29.44	32.04	30.32	28.72	27.42
	0.743	1.074	1.411	1.911	1.330	1.607	2.315	2.867
RORD-WMF	35.49	33.52	31.61	29.70	32.45	30.73	29.07	27.37
	0.598	0.960	1.287	1.762	1.187	1.574	2.230	2.884
DWM	34.44	32.51	30.70	28.74	31.55	29.77	28.12	26.40
	0.838	1.546	1.748	2.737	1.510	1.908	2.784	3.233
DARD-BF	35.14	33.08	31.16	29.18	32.21	30.23	28.51	27.20
	0.703	1.120	1.524	2.014	1.286	1.715	2.446	2.997
SG-DARD-TRIF	35.23	33.23	31.37	29.45	32.32	30.34	28.67	27.39
	0.651	1.005	1.389	1.903	1.218	1.604	2.378	2.876