Many Local Pattern Texture Features: Which Is Better for Image-Based Multilabel Human Protein Subcellular Localization Classification?
Table 1
Comparison SDA results derived from the combination of different local pattern features and global features.
DB
Fold
1096 dimensional SLFs_LBP feature fed into SDA
1040 dimensional SLFs_CLBP feature fed into SDA
1607 dimensional SLFs_LTrP feature fed into SDA
db1
fold 1
55 (D1-H30-L24)
44%
50 (D1-H25-C24)
48%
69 (D2-H21-T46)
67%
fold 2
48 (D3-H22-L23)
48%
40 (D2-H19-C19)
48%
59 (D3-H22-T34)
58%
db2
fold 1
66 (D1-H39-L26)
39%
49 (D1-H32-C16)
33%
70 (D2-H29-T39)
56%
fold 2
49 (D3-H21-L25)
51%
42 (D3-H23-C16)
38%
52 (D2-H17-T33)
63%
db3
fold 1
58 (D2-H28-L28)
48%
50 (D1-H27-C22)
44%
66 (D2-H20-T44)
67%
fold 2
34 (D3-H13-L18)
53%
44 (D3-H23-C18)
41%
45 (D2-H12-T31)
69%
db4
fold 1
53 (D1-H28-L24)
45%
53 (D2-H27-C24)
45%
69 (D22-H3-T44)
64%
fold 2
39 (D1-H16-L22)
56%
42 (D2-H24-C16)
38%
48 (D1-H16-T31)
65%
db5
fold 1
53 (D1-H25-L27)
51%
49 (D2-H27-C20)
41%
64 (D1-H24-T39)
61%
fold 2
44 (D3-H15-L26)
59%
42 (D3-H19-C20)
48%
46 (D1-H11-T34)
74%
db6
fold 1
49 (D1-H22-L26)
53%
46 (D2-H23-C21)
46%
68 (D3-H20-T45)
66%
fold 2
40 (D3-H22-L15)
38%
46 (D3-H27-C16)
35%
45 (D1-H13-T31)
66%
db7
fold 1
56 (D1-H32-L23)
41%
49 (D1-H28-C10)
20%
65 (D3-H19-T43)
66%
fold 2
39 (D3-H18-L18)
46%
41 (D2-H17-C22)
54%
42 (D1-H12-T29)
69%
db8
fold 1
53 (D1-H26-L26)
49%
44 (D2-H23-C19)
43%
61 (D1-H17-T43)
70%
fold 2
40 (D2-H14-L24)
60%
42 (D2-H21-C19)
45%
51 (D1-H13-T37)
73%
db9
fold 1
54 (D1-H29-L24)
44%
51 (D3-H31-C17)
33%
67 (D3-H23-T41)
61%
fold 2
43 (D2-H17-L24)
56%
43 (D3-H23-C17)
40%
54 (D2-H16-T36)
67%
db10
fold 1
55 (D1-H24-L30)
55%
45 (D1-H25-C19)
42%
65 (D2-H23-T40)
62%
fold 2
46 (D2-H17-L27)
59%
47 (D3-H19-C24)
51%
52 (D2-H15-T35)
67%
D, H, L, C, and T denote DNA-protein overlap feature, Haralick feature, LBP, CLBP, and LTrP, respectively. DB denotes 10 different lengths of vanishing moment of Daubechies wavelet. All percentages denote the proportion of local pattern features in the whole feature set, and boldface type denotes the value closest to the mean of that column.