Table 2.

Voxel-and object-based classifications of benchmark set 2 for tomograms at four different SNR levels

	Initial		GHMRF
SNR	Precision	Recall	Precision	Recall
	Voxel-based classification
∞	0.50 (0.16)	0.72 (0.17)	0.72 (0.14)	0.84 (0.16)
5	0.45 (0.18)	0.68 (0.19)	0.72 (0.18)	0.79 (0.20)
1	0.30 (0.18)	0.53 (0.18)	0.52 (0.21)	0.57 (0.22)
0.5	0.31 (0.15)	0.54 (0.16)	0.51 (0.19)	0.56 (0.20)
	Object-based classification
∞	0.85 (0.19)	0.85 (0.19)	0.91 (0.12)	0.91 (0.12)
5	0.79 (0.22)	0.80 (0.22)	0.87 (0.15)	0.87 (0.15)
1	0.61 (0.23)	0.61 (0.23)	0.70 (0.17)	0.71 (0.17)
0.5	0.58 (0.20)	0.59 (0.20)	0.64 (0.16)	0.65 (0.16)

For each SNR level, 50 tomograms are generated. Each tomogram consists of 40 complexes of four different types. Values are the mean precision and recall for the set of 50 classifications with SDs shown in brackets.