Total Variation Regularization Algorithms for Images Corrupted with Different Noise Models: A Review

Table 5

Denoising performance results for the mixed Gaussian Impulse TV method for the grayscale “Peppers” and the color “Lena” images (both 512×512). The Gaussian plus Salt & Pepper noise case is marked as “G + S & P”, whereas the Gaussian plus random-valued impulse noise case is marked as “G + R.”

Image	Noise			SNR (dB)		SSIM		Time (s)
	𝜎	𝑝S&P	𝑝R	G + S & P	G + R	G + S & P	G + R	G + S & P	G + R
Peppers (gray)	5/255	0.3	0.1	19.88	18.80	0.87	0.83	10.43	12.05
		0.5	0.2	18.15	16.43	0.85	0.76	10.85	11.03
		0.7	0.3	15.99	14.62	0.82	0.70	11.95	10.81
	10/255	0.3	0.1	18.28	18.13	0.82	0.81	11.28	12.22
		0.5	0.2	16.97	16.68	0.81	0.78	11.52	12.49
		0.7	0.3	14.89	15.33	0.78	0.75	12.94	11.89
	15/255	0.3	0.1	17.02	17.37	0.79	0.78	11.45	12.52
		0.5	0.2	15.98	16.26	0.78	0.76	11.81	12.52
		0.7	0.3	13.98	15.09	0.75	0.73	13.17	12.04
Lena (color)	5/255	0.3	0.1	21.16	17.38	—	—	38.09	30.62
		0.5	0.2	19.76	15.13	—	—	37.57	29.06
		0.7	0.3	17.51	13.48	—	—	42.24	29.31
	10/255	0.3	0.1	18.93	15.04	—	—	38.18	30.50
		0.5	0.2	17.85	13.70	—	—	40.10	31.10
		0.7	0.3	16.02	12.68	—	—	44.05	29.46
	15/255	0.3	0.1	17.76	12.78	—	—	39.08	30.19
		0.5	0.2	16.77	12.15	—	—	40.82	29.18
		0.7	0.3	15.11	11.41	—	—	45.10	32.65