1 Introduction
2 Materials and methods
2.1 Differential excitation (DE)
2.2 Gradient orientation (GO)
2.3 Basic WLD
-
T, the number of dominant orientations ϕ t : t = 0, 1, 2, …, T−1,
-
M, the number of segments H m,t of each sub-histogram H t corresponding to a dominant orientation ϕ t , and
-
S, the number of bins in each sub-histogram H m,t .
2.4 Multiscale WLD
2.5 Spatial WLD
2.6 Multiscale spatial WLD
2.7 Significance of features
2.8 Support vector machine (SVM)
3 False-positive reduction system
3.1 Database
4 Experiments and discussion
4.1 Evaluation strategy
4.2 Optimization of parameters
4.2.1 Effect of T, M, and S
Number blocks | (T, M, S) | Sensitivity | Specificity | Accuracy | Az |
---|---|---|---|---|---|
4 × 4 | (4, 4, 5) |
98.45 ± 1.33
|
97.56 ± 1.23
|
98.00 ± 0.56
|
0.98 ± 0.006
|
(12, 4, 20) | 98.02 ± 1.59 | 96.68 ± 0.66 | 97.36 ± 0.74 | 0.97 ± 0.009 | |
5 × 5 | (4, 4, 5) | 98.25 ± 0.85 | 97.45 ± 0.83 | 97.85 ± 0.76 | 0.97 ± 0.008 |
(12, 4, 20) | 97.88 ± 2.17 | 96.28 ± 2.24 | 97.07 ± 1.97 | 0.97 ± 0.02 |
4.2.2 Effect of scales (P, R)
4.2.3 Effect of number of blocks and feature selection
Number blocks | Number features | Sensitivity | Specificity | Accuracy | Az | (C, γ) | (σ, λ) |
---|---|---|---|---|---|---|---|
4 × 4 | 1280 | 98.45 ± 1.33 | 97.56 ± 1.23 | 98.00 ± 0.56 | 0.98 ± 0.006 | (29, 2−17) | |
220 (A. F. S.) | 99.02 ± 0.47 | 98.14 ± 0.39 | 98.58 ± 0.26 | 0.98 ± 0.002 | (29, 2−17) | (0.3, 0.7) | |
5 × 5 | 2000 | 98.25 ± 0.85 | 97.45 ± 0.83 | 97.85 ± 0.76 | 0.97 ± 0.008 | (29, 2−17) | |
261 (A. F. S.) |
98.82 ± 0.44
|
99.03 ± 0.81
|
98.93 ± 0.56
|
0.99 ± 0.003
| (29, 2−17) | (0.1, 0.5) |
4.3 Discussion
4.4 Comparison
Sensitivity | Specificity | Accuracy | Az | |
---|---|---|---|---|
MSWLD |
98.82 ± 0.44
|
99.03 ± 0.81
|
98.93 ± 0.56
|
0.99 ± 0.003
|
LBP | 90.35 ± 3.18 | 93.66 ± 1.18 | 92.00 ± 0.99 | 0.92 ± 0.016 |
Basic WLD | 75.62 ± 6.78 | 69.22 ± 8.60 | 72.29 ± 2.39 | 0.72 ± 0.024 |