Table 3 Comparison of the classification accuracy of several microstructural representations on dataset 2.

From: Compact representations of microstructure images using triplet networks

Method

Dimensionality

Out-of-bag acc. [%]

Test acc. [%]

Haralick5

13

98.4 ± 0.3

91.5 ± 3.2

lbp5

20

92.0 ± 1.0

83.6 ± 4.2

Greyco5

4

63.0 ± 1.0

50.9 ± 1.0

Correlations8

20

79.8 ± 1.0

69.8 ± 2.6

Surf15

100

28.1 ± 1.4

28.4 ± 1.2

VGG16 VLAD C437

16,384

65.5 ± 3.8

77.3 ± 0.8

VGG16 mean C437

512

99.4 ± 0.1

97.8 ± 0.5

VGG16 max C437

512

98.6 ± 0.2

96.1 ± 2.2

Morphological6

21

72.0 ± 1.6

66.7 ± 0.7

Texture26

8

89.6 ± 1.9

83.4 ± 2.2

Morph. + texture6

29

93.3 ± 1.1

88.9 ± 1.8

Morph. + texture + ga12

16

92.7 ± 1.9

87.5 ± 1.8

Triplets

2

71.7 ± 4.8

66.2 ± 6.5

Triplets

3

83.7 ± 3.9

78.9 ± 5.3

Triplets

10

97.7 ± 0.6

94.6 ± 2.0

Triplets

512

99.9 ± 0.1

99.5 ± 0.4

  1. Both the accuracy (acc.) for the test data and the out-of-bag accuracy are listed. All accuracies were obtained using a random forest classifier. We also list the SDs, which are computed by using threefold cross-validation. Even on unseen materials the proposed methods perform as well as the other representations with similar dimensionality.
Back to article page