Improved DCT-Based Nonlocal Means Filter for MR Images Denoising

Table 1

Comparisons of experimental results in PSNR.

Test images	Algorithms	Noise level (dB)
		3%	6%	9%	12%	15%	18%
T1-weighted MR image	Noisy	29.86	24.3	21.26	19.56	18.35	17.44
	NLM (h)	32.21 (3^4.38)	26.73 (6^2.94)	23.44 (9^2.46)	21.14 (12^2.24)	19.61 (15^2.12)	18.36 (18^2.04)
	UNLM (h)	32.8 (3^4.38)	27.46 (6^2.94)	24.14 (9^2.46)	21.9 (12^2.3)	20.3 (15^2.12)	19.02 (18^2.06)
	UNLM-DCT (h,d)	33.02 (3^4,17)	27.92(6^2.48,12)	24.9 (9^2.08,12)	22.76 (12^1.86,12)	21.05 (15^1.74,12)	20.16 (18^1.26,3)
T2-weighted MR image	Noisy	30.48	25.21	22.44	20.66	19.4	18.45
	NLM (h)	32.3 (3^4.62)	27.3 (6^3.02)	24.32 (9^2.54)	22.27 (12^2.3)	20.8 (15^2.18)	19.68 (18^2.06)
	UNLM (h)	32.7 (3^4.62)	27.79 (6^3.02)	24.83 (9^2.62)	22.8 (12^2.36)	21.36 (15^2.18)	20.19 (18^2.06)
	UNLM-DCT (h,d)	32.77 (3^4.32,25)	28.13 (6^2.64,13)	25.53 (9^2.16,11)	23.73 (12^1.86,8)	22.49 (15^1.56,4)	21.56 (18^1.44,3)
PD-weighted MR image	Noisy	30.52	25.23	22.43	20.61	19.31	18.33
	NLM (h)	32.9 (3^4.54)	27.98 (6^3.02)	24.87 (9^2.54)	22.7 (12^2.36)	20.97 (15^2.18)	19.63 (18^2.12)
	UNLM (h)	33.36 (3^4.54)	28.59 (6^3.02)	25.55 (9^2.62)	23.37 (12^2.36)	21.57 (15^2.24)	20.29 (18^2.12)
	UNLM-DCT (h,d)	33.58 (3^4.16, 24)	28.64 (6^2.72, 24)	25.76 (9^1.92, 6)	23.81 (12^1.74, 6)	22.41 (15^1.44, 3)	21.31 (18^1.32, 3)