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

Table 1

Deconvolution performance results for the standard ℓ²-TV and for the nonnegative ℓ²-TV methods for the grayscale “Goldhill” (720×576) image and the color “Lena” (512×512) image.

Image	Noise (𝜎^2)	SNR (dB)		SSIM		Time (s)
		ℓ^2-TV	Nonneg.	ℓ^2-TV	Nonneg.	ℓ^2-TV	Nonneg.
Goldhill (gray)	10^−2	13.39	13.89	0.59	0.60	6.69	11.99
	4⋅10^−2	11.59	12.12	0.55	0.54	7.61	12.05
Lena (color)	10^−2	14.98	15.17	—	—	11.21	23.34
	4⋅10^−2	13.02	13.15	—	—	12.50	23.19