1 Introduction
Acronym | Definition |
---|---|
BIF | Biologically-inspired features (Song and Tao 2010) |
BOP | Bag-of-parts (Juneja et al. 2013) |
BoW | Bag-of-words (Fei-Fei et al. 2007a) |
CENTRIST | Census transform histogram (Wu and Rehg 2011) |
CNN | |
DeCAF | Deep convolutional activation feature (Donahue et al. 2014) |
GIST | An abstract representation of the scene (Oliva and Torralba 2001) |
HIK | Histogram intersection kernel (Maji et al. 2008) |
HMAX | Hierarchical model and X (Riesenhuber and Poggio 1999) |
HOG | Histogram of oriented gradients (Dalal and Triggs 2005) |
HOG-SPM | HOG with spatial pyramid matching (Lazebnik et al. 2006) |
HSOG | Histogram of the second-order gradients (Huang et al. 2014a) |
LBP | Local binary pattern (Ojala et al. 1996) |
LBP-HF | LBP with Fourier histogram (Ahonen et al. 2009) |
LCS | Local color statistic (Clinchant et al. 2007) |
OB | Object bank (Li et al. 2010) |
PLBP | LBP with pyramid representation (Qian et al. 2011) |
RBF | Radial basis function kernel (Scholkopf et al. 1997) |
SEV | Sequential edge vectors (Morikawa and Shibata 2012) |
SIFT | Scale-invariant feature transform (Lowe 2004) |
SIFT-FV | SIFT with Fisher vector (Sanchez et al. 2013) |
SIFT-LLC | SIFT with locality-constrained linear coding (Wang et al. 2010) |
SIFT-ScSPM | SIFT with sparse coding based spatial pyramid matching (Yang et al. 2009) |
SIFT-SPM | SIFT with spatial pyramid matching (Lazebnik et al. 2006) |
SURF | Speeded up robust features (Bay et al. 2008) |
SVM | Support vector machine (Burges 1998) |
2 Visual descriptors
Approaches | Representative works |
---|---|
Biologically-inspired feature extraction | |
Visual cell model | HMAX (Serre et al. 2005) |
Layout properties | GIST (Oliva and Torralba 2001) |
Deep learning features | Convolutional neural networks (Zhou et al. 2014) |
OverFeat (Sermanet et al. 2014) | |
DeCAF Donahue et al. (2014) | |
Local feature extraction | |
Patch-based features | SIFT (Lowe 2004) |
HOG (Dalal and Triggs 2005) | |
LBP (Ojala et al. 1996) | |
CENTRIST (Wu and Rehg 2011) | |
HSOG (Huang et al. 2014a) | |
Object-based model | Object bank (Li et al. 2010) |
Region-based model | Edge vectors (Morikawa and Shibata 2012) |
Bag of parts (Juneja et al. 2013) | |
Keypoint-based features | SURF (Bay et al. 2008) |
FREAK (Alahi et al. 2012) | |
BRISK (Leutenegger et al. 2011) | |
Global feature formation | |
Principal component analysis | SIFT features with PCA (Ke and Sukthankar 2004) |
Histogram | Multi-resolution histogram (Hadjidemetriou et al. 2004) |
Feature encoding | SPM (Lazebnik et al. 2006) |
ScSPM (Yang et al. 2009) | |
LLC (Wang et al. 2010) | |
2.1 Biologically-inspired feature extraction models
2.1.1 HMAX model
2.1.2 GIST model
2.1.3 Deep learning
2.2 Local feature extraction
2.2.1 Patch-based local features
2.2.2 Object-based local features
2.2.3 Region-based local features
2.3 Global feature formation
2.3.1 Principal component analysis
2.3.2 Histogram
2.3.3 Bag-of-words
2.3.4 Fisher vector
2.3.5 Composite global features
3 Data sets and performance measures
3.1 Data sets for scene categorization
Data set | Data set size (images) | Image categories |
---|---|---|
8-outdoor-scene (Oliva and Torralba 2001) | 2600 | 8 outdoor scenes |
13-natural-scene (Fei-Fei and Perona 2005) | 3759 | 13 natural scenes |
15-scene (Lazebnik et al. 2006) | 4485 | 15 indoor/outdoor scenes |
67-indoor-scene (Quattoni and Torralba 2009) | 15,620 | 67 indoor scenes |
SUN397 (Xiao et al. 2014a) | 108,754 | 397 general scenes |
Places205 (Zhou et al. 2014) | 2,448,873 | 205 general scenes |
3.2 Performance measures
-
True positives (tp) is the number of test samples in the positive class that are correctly classified.
-
False positives (fp) is the number of test samples in the negative class that are incorrectly classified.
-
False negatives (fn) is the number of test samples in the positive class that are incorrectly classified.
-
True negatives (tn) is the number of test samples in the negative class that are correctly classified.
4 Experimental evaluation and results
4.1 Image data sets
4.2 Implementation of visual descriptors
4.3 Classification procedure
4.4 Classification results
4.4.1 Classification results on the 15-scene data set
ID | Algorithms | CR (%) | Precision (%) | Recall (%) | F-measure (%) | AUC (%) |
---|---|---|---|---|---|---|
1 | SIFT-ScSPM |
\( \mathbf {84.5 \pm 1.5} \)
|
\(\mathbf { 84.8 \pm 1.7 }\)
|
\( \mathbf {84.0\pm 1.4}\)
|
\( \mathbf {84.1 \pm 1.4} \)
|
\(\mathbf {98.9 \pm 0.1}\)
|
2 | SIFT-LLC |
\( 83.0\pm 1.3 \)
|
\( 83.0\pm 1.5\)
|
\( 82.2 \pm 1.5\)
|
\( 82.3 \pm 1.4 \)
|
\( 98.8 \pm 0.2\)
|
3 | SIFT-FV |
\( 80.2 \pm 1.8 \)
|
\( 79.6 \pm 1.9\)
|
\( 79.2 \pm 1.9 \)
|
\( 79.0 \pm 2.0 \)
|
\( 98.4 \pm 0.1\)
|
4 | HOG-SPM |
\( 70.6 \pm 2.7 \)
|
\(71.7 \pm 2.7\)
|
\(68.7 \pm 2.6\)
|
\(70.0 \pm 2.4\)
|
\(96.3 \pm 0.3\)
|
5 | OB |
\( 79.9\pm 1.6\)
|
\(80.0 \pm 1.6\)
|
\( 79.1\pm 2.0\)
|
\( 79.0 \pm 1.9\)
|
\(97.7 \pm 0.3 \)
|
6 | SIFT-SPM |
\( 60.9\pm 4.0 \)
|
\(62.3\pm 5.6\)
|
\(57.8 \pm 4.3 \)
|
\(57.0 \pm 4.7\)
|
\(94.4 \pm 0.5 \)
|
7 | SURF-ScSPM |
\( 73.3 \pm 2.1 \)
|
\( 72.9 \pm 2.1\)
|
\( 72.3 \pm 2.3 \)
|
\(72.3 \pm 2.4\)
|
\( 97.3\pm 0.4 \)
|
8 | GIST |
\( 71.5 \pm 1.2 \)
|
\(71.1 \pm 1.4 \)
|
\(70.6 \pm 1.5\)
|
\(71.2 \pm 1.0\)
|
\( 95.7\pm 0.4\)
|
9 | CENTRIST |
\( 72.7 \pm 1.4 \)
|
\(72.8 \pm 1.1\)
|
\(72.2 \pm 1.9 \)
|
\(71.9 \pm 1.7 \)
|
\( 95.9 \pm 0.3\)
|
10 | LBP |
\( 71.1 \pm 2.9 \)
|
\(69.8 \pm 4.0 \)
|
\(70.3 \pm 3.4\)
|
\(69.3 \pm 3.9\)
|
\( 94.9 \pm 1.2\)
|
11 | Uniform LBP |
\( 56.2 \pm 3.0 \)
|
\(55.6 \pm 5.8\)
|
\(54.7 \pm 3.2\)
|
\(52.6 \pm 3.6\)
|
\(92.8 \pm 1.0\)
|
12 | LBP-HF |
\( 64.9 \pm 3.4 \)
|
\(63.8\pm 4.6\)
|
\(63.9 \pm 3.8 \)
|
\(62.5 \pm 4.3\)
|
\( 92.8 \pm 1.0\)
|
13 | PLBP |
\( 53.8 \pm 3.6 \)
|
\(52.7 \pm 4.7\)
|
\( 52.4\pm 3.6\)
|
\( 51.5 \pm 4.1\)
|
\( 92.1 \pm 1.5\)
|
14 | HMAX |
\(61.1 \pm 4.1\)
|
\( 60.8 \pm 4.0\)
|
\( 59.8 \pm 4.5 \)
|
\( 59.9 \pm 4.3 \)
|
\( 79.7 \pm 6.9\)
|
ID | Algorithms | CR (%) | Precision (%) | Recall (%) | F-measure (%) | AUC (%) |
---|---|---|---|---|---|---|
1 | SIFT-ScSPM |
\( \mathbf {83.8 \pm 1.7} \)
|
\( \mathbf {84.2 \pm 1.6} \)
|
\( \mathbf {83.2 \pm 1.6}\)
|
\( \mathbf {83.5 \pm 1.3 }\)
|
\( \mathbf {98.3 \pm 0.0} \)
|
2 | SIFT-LLC |
\( 82.2\pm 1.1 \)
|
\( 82.5 \pm 1.2 \)
|
\( 81.5 \pm 1.1 \)
|
\( 81.5 \pm 1.2\)
|
\(98.2 \pm 0.1\)
|
3 | SIFT-FV |
\( 79.4\pm 1.7 \)
|
\( 78.9 \pm 2.1\)
|
\( 78.1 \pm 2.1 \)
|
\( 77.6 \pm 2.3 \)
|
\( 97.5 \pm 0.2\)
|
4 | HOG-SPM |
\( 78.9 \pm 1.2 \)
|
\(76.8 \pm 3.2\)
|
\(76.3 \pm 2.9\)
|
\(76.1 \pm 3.1\)
|
\(96.5 \pm 0.5\)
|
5 | OB |
\( 73.2\pm 2.0\)
|
\( 73.5\pm 1.9\)
|
\( 72.0\pm 2.4\)
|
\(72.1 \pm 2.3\)
|
\(95.5 \pm 0.4 \)
|
6 | SIFT-SPM |
\( 72.9 \pm 1.3 \)
|
\( 72.0\pm 1.6\)
|
\(71.6 \pm 1.4\)
|
\(71.3 \pm 1.4\)
|
\(95.2 \pm 0.4\)
|
7 | SURF-ScSPM |
\( 72.4 \pm 1.6\)
|
\(71.6 \pm 1.8\)
|
\(71.1 \pm 1.7\)
|
\( 70.6\pm 1.8\)
|
\( 95.8 \pm 0.4 \)
|
8 | GIST |
\( 72.8 \pm 1.2 \)
|
\(72.2 \pm 1.0 \)
|
\(71.8 \pm 1.4\)
|
\(71.5 \pm 1.2\)
|
\(95.2 \pm 0.2\)
|
9 | CENTRIST |
\( 72.6 \pm 1.8 \)
|
\(72.7 \pm 1.1 \)
|
\(72.0 \pm 2.1\)
|
\(71.8 \pm 1.8\)
|
\(91.2 \pm 1.0\)
|
10 | LBP |
\( 70.9 \pm 4.2 \)
|
\(69.6\pm 5.2\)
|
\(70.1 \pm 4.8\)
|
\(69.2 \pm 5.3\)
|
\(96.2 \pm 0.7\)
|
11 | Uniform LBP |
\( 67.8 \pm 3.9 \)
|
\(66.8 \pm 5.0\)
|
\(66.8 \pm 4.4\)
|
\(65.6 \pm 4.9\)
|
\(94.0 \pm 0.8\)
|
12 | LBP-HF |
\( 67.3 \pm 4.4 \)
|
\(66.0 \pm 5.4\)
|
\( 66.5 \pm 4.9\)
|
\(66.3 \pm 4.4\)
|
\(94.0 \pm 0.8\)
|
13 | PLBP |
\( 69.4 \pm 3.5 \)
|
\( 69.2 \pm 4.9\)
|
\( 68.2 \pm 4.1\)
|
\( 68.1\pm 4.4\)
|
\( 95.4\pm 1.0\)
|
14 | HMAX |
\(62.4 \pm 3.7 \)
|
\( 61.6 \pm 3.4\)
|
\( 61.0 \pm 4.0\)
|
\( 60.8 \pm 3.7 \)
|
\( 79.5 \pm 6.0 \)
|
ID | Algorithms | CR (%) | Precision (%) | Recall (%) | F-measure (%) | AUC (%) |
---|---|---|---|---|---|---|
1 | SIFT-ScSPM |
\( \mathbf { 83.6 \pm 1.6} \)
|
\( \mathbf {84.2\pm 1.7} \)
|
\( \mathbf {83.2 \pm 1.4}\)
|
\( \mathbf {83.2 \pm 1.4} \)
|
\( \mathbf {98.2\pm 0.1} \)
|
2 | SIFT-LLC |
\( 82.6 \pm 1.5 \)
|
\( 82.8 \pm 1.4 \)
|
\( 82.1 \pm 1.3 \)
|
\( 82.1 \pm 1.3\)
|
\(98.2 \pm 0.0\)
|
3 | SIFT-FV |
\( 80.0 \pm 1.7 \)
|
\( 80.1\pm 2.0\)
|
\( 79.7 \pm 1.4 \)
|
\( 79.3 \pm 1.7 \)
|
\( 97.7 \pm 0.2\)
|
4 | HOG-SPM |
\( 80.0 \pm 2.6 \)
|
\(79.5 \pm 2.5\)
|
\(79.7 \pm 2.5 \)
|
\(79.6 \pm 2.4\)
|
\( 97.8 \pm 0.2 \)
|
5 | OB |
\( 76.9 \pm 2.0\)
|
\( 76.9 \pm 1.8\)
|
\( 76.0\pm 2.3\)
|
\(76.0 \pm 2.2\)
|
\( 97.4\pm 0.2 \)
|
6 | SIFT-SPM |
\( 77.9 \pm 1.2 \)
|
\(77.5\pm 1.3\)
|
\(76.8 \pm 1.0 \)
|
\(76.8 \pm 1.0\)
|
\(97.0 \pm 0.3\)
|
7 | SURF-ScSPM |
\( 72.3\pm 2.4\)
|
\(71.4 \pm 2.5\)
|
\(71.2 \pm 2.5\)
|
\( 70.6 \pm 2.6\)
|
\(95.6 \pm 0.4\)
|
8 | GIST |
\( 72.1 \pm 0.7 \)
|
\( 71.8 \pm 0.7 \)
|
\(71.4 \pm 0.9\)
|
\(72.1 \pm 0.9\)
|
\( 95.2 \pm 0.2 \)
|
9 | CENTRIST |
\( 70.9 \pm 2.4 \)
|
\(71.3 \pm 1.8 \)
|
\(70.1 \pm 2.6 \)
|
\(70.0 \pm 2.3\)
|
\(95.9 \pm 0.4 \)
|
10 | LBP |
\( 71.9 \pm 2.5 \)
|
\(71.0 \pm 3.2\)
|
\(71.1 \pm 3.2 \)
|
\(70.5 \pm 3.4\)
|
\(95.7 \pm 0.8 \)
|
11 | Uniform LBP |
\( 70.6 \pm 3.5 \)
|
\( 70.4\pm 4.4\)
|
\(69.8 \pm 4.0\)
|
\(69.3 \pm 4.5 \)
|
\( 95.7\pm 1.0\)
|
12 | LBP-HF |
\( 66.4 \pm 3.3 \)
|
\(65.6\pm 4.4\)
|
\(65.5 \pm 3.7\)
|
\(64.8 \pm 4.2\)
|
\( 95.7\pm 1.0\)
|
13 | PLBP |
\( 73.0 \pm 3.4 \)
|
\( 72.3\pm 5.1\)
|
\(72.9 \pm 4.2\)
|
\(72.6 \pm 4.7\)
|
\( 96.6 \pm 1.0\)
|
14 | HMAX |
\(63.9 \pm 3.4 \)
|
\(63.5 \pm 3.6\)
|
\( 62.5 \pm 3.7\)
|
\(62.6\pm 3.6 \)
|
\( 82.9\pm 8.3 \)
|
ID | Algorithms | CR (%) | Precision (%) | Recall (%) | F-measure (%) | AUC (%) |
---|---|---|---|---|---|---|
1 | SIFT-ScSPM (linear) |
\( \mathbf {89.8 \pm 1.8}\)
|
\( \mathbf {90.4\pm 1.8 }\)
|
\( \mathbf {90.0 \pm 2.2}\)
|
\(\mathbf { 90.0 \pm 2.1} \)
|
\( \mathbf {98.4 \pm 0.4}\)
|
2 | SIFT-LLC (linear) |
\( 88.1 \pm 2.1 \)
|
\( 88.6\pm 2.0 \)
|
\( 88.3 \pm 2.6 \)
|
\( 88.3 \pm 2.4 \)
|
\(98.2 \pm 0.6\)
|
3 | SIFT-FV (linear) |
\( 88.1\pm 3.3 \)
|
\(88.2\pm 3.4\)
|
\( 88.0 \pm 3.9 \)
|
\( 87.9 \pm 3.8\)
|
\( 98.7 \pm 0.4\)
|
4 | HOG-SPM(HIK) |
\( 88.1 \pm 2.5 \)
|
\(89.0 \pm 2.1\)
|
\(88.2 \pm 3.2\)
|
\(88.3 \pm 2.9 \)
|
\( 98.3\pm 0.6 \)
|
5 | OB (linear) |
\(87.5 \pm 2.5 \)
|
\(88.0 \pm 2.3 \)
|
\( 87.6 \pm 3.1\)
|
\(87.5 \pm 2.8\)
|
\(98.4\pm 0.5 \)
|
6 | SIFT-SPM(HIK) |
\( 87.4 \pm 2.3 \)
|
\(87.9 \pm 2.2\)
|
\(87.8 \pm 2.5 \)
|
\(87.6 \pm 2.4\)
|
\(98.1 \pm 0.4\)
|
7 | SURF-ScSPM (linear) |
\( 85.9 \pm 2.6 \)
|
\( 85.4 \pm 2.8 \)
|
\(85.0 \pm 3.4\)
|
\(85.9 \pm 3.2\)
|
\(98.0 \pm 0.5\)
|
8 | GIST (RBF) |
\( 85.3 \pm 2.5 \)
|
\(85.9 \pm 2.4 \)
|
\(85.4 \pm 3.1\)
|
\( 85.4 \pm 2.9 \)
|
\(98.0 \pm 0.8\)
|
9 | CENTRIST (linear) |
\( 83.5 \pm 4.9 \)
|
\( 84.3 \pm 4.8 \)
|
\( 83.6 \pm 5.8\)
|
\(83.5 \pm 5.5 \)
|
\( 97.2 \pm 1.1\)
|
10 | LBP (HIK) |
\( 76.6 \pm 4.3 \)
|
\(77.0 \pm 4.6 \)
|
\(76.8 \pm 5.1\)
|
\(76.5 \pm 4.9\)
|
\(95.8 \pm 1.2\)
|
11 | Uniform LBP (HIK) |
\( 75.2 \pm 4.6 \)
|
\( 76.0 \pm 4.6\)
|
\(75.2 \pm 5.8 \)
|
\( 74.9\pm 5.5 \)
|
\(95.0 \pm 1.6\)
|
12 | LBP-HF (RBF) |
\(71.9\pm 3.7 \)
|
\(72.1 \pm 4.5\)
|
\( 71.9 \pm 4.7 \)
|
\(71.4 \pm 4.6\)
|
\(92.5 \pm 2.6\)
|
13 | PLBP (HIK) |
\( 81.4 \pm 2.9\)
|
\( 81.6\pm 3.6\)
|
\( 81.3 \pm 3.7 \)
|
\(81.2 \pm 3.8\)
|
\( 96.7\pm 1.1\)
|
14 | HMAX (HIK) |
\( 79.8 \pm 0.8\)
|
\( 80.1 \pm 0.7\)
|
\( 80.2 \pm 0.7\)
|
\(80.0 \pm 0.7 \)
|
\( 96.5 \pm 0.3\)
|
4.4.2 Classification results on the 8-outdoor-scene data set
4.4.3 Classification results on the 67-indoor-scene data set
ID | Algorithms | CR (%) | Precision (%) | Recall (%) | F-measure (%) | AUC (%) |
---|---|---|---|---|---|---|
1 | SIFT-ScSPM (linear) |
\( \mathbf { 45.6 \pm 1.0}\)
|
\(\mathbf { 45.4 \pm 2.5}\)
|
\(\mathbf { 35.8 \pm 1.5}\)
|
\( \mathbf { 36.9\pm 1.7}\)
|
\( \mathbf {91.7\pm 0.4}\)
|
2 | SIFT-LLC (linear) |
\( 43.9\pm 0.4\)
|
\( 44.7\pm 1.2\)
|
\( 35.2 \pm 0.8 \)
|
\( 36.3 \pm 0.9\)
|
\( 91.3 \pm 0.5\)
|
3 | SIFT-FV (linear) |
\( 41.5\pm 1.2 \)
|
\( 46.0\pm 1.3 \)
|
\( 31.4 \pm 1.0 \)
|
\( 32.5 \pm 1.0\)
|
\( 90.4 \pm 0.6\)
|
4 | HOG-SPM (HIK) |
\( 31.0 \pm 0.5 \)
|
\( 27.5 \pm 0.8\)
|
\( 25.5\pm 0.6\)
|
\(25.8 \pm 0.6\)
|
\( 85.1\pm 0.6\)
|
5 | OB (linear) |
\( 45.3 \pm 0.5\)
|
\( 43.7\pm 1.0 \)
|
\( 39.4\pm 0.7\)
|
\(39.2 \pm 0.7\)
|
\( 91.7 \pm 0.4\)
|
6 | SIFT-SPM (HIK) |
\( 31.4 \pm 0.8 \)
|
\( 28.4 \pm 1.1\)
|
\( 25.9\pm 0.7\)
|
\(26.4 \pm 0.8\)
|
\( 85.3\pm 0.5\)
|
7 | SURF-ScSPM (linear) |
\(34.5 \pm 0.7\)
|
\(35.5 \pm 0.9\)
|
\( 25.4 \pm 0.4\)
|
\( 26.2\pm 0.2\)
|
\( 87.6\pm 0.3\)
|
8 | GIST (RBF) |
\( 30.9\pm 0.9\)
|
\( 28.0\pm 0.1\)
|
\(25.2 \pm 0.3\)
|
\(25.7 \pm 0.4\)
|
\( 84.4\pm 0.4\)
|
9 | CENTRIST (linear) |
\( 12.2\pm 11.0\)
|
\(15.4 \pm 8.5\)
|
\( 10.8\pm 9.6\)
|
\(10.1 \pm 9.6\)
|
\(78.4 \pm 3.8\)
|
10 | LBP (HIK) |
\( 22.9 \pm 0.6\)
|
\(21.0 \pm 1.0 \)
|
\( 17.7\pm 0.9 \)
|
\( 18.4 \pm 0.9\)
|
\( 81.7\pm 0.5\)
|
11 | Uniform LBP (HIK) |
\( 22.0 \pm 1.1\)
|
\(18.5 \pm 0.4\)
|
\( 16.2\pm 1.0\)
|
\( 16.6\pm 0.9\)
|
\( 80.0\pm 0.6\)
|
12 | LBP-HF (RBF) |
\( 15.4\pm 1.2\)
|
\( 12.7\pm 0.8 \)
|
\( 11.9\pm 0.7\)
|
\( 11.8 \pm 0.7\)
|
\( 76.3\pm 1.0\)
|
13 | PLBP (HIK) |
\( 27.2\pm 1.0\)
|
\( 24.1\pm 1.3\)
|
\( 21.6 \pm 1.0 \)
|
\(22.2 \pm 1.0\)
|
\( 84.5 \pm 0.3\)
|
14 | HMAX (HIK) |
\( 11.6 \pm 2.3\)
|
\(10.5 \pm 2.7\)
|
\( 9.4\pm 1.9 \)
|
\( 9.5\pm 2.1\)
|
\( 72.9 \pm 3.0\)
|
4.4.4 Classification results on the SUN397 data set
ID | Algorithms | 8-outdoor-scene | 15-scene | 67-indoor-scene | SUN397 |
---|---|---|---|---|---|
1 | SIFT-ScSPM | 1.8827 | 1.3005 | 1.0364 | 1.0064 |
2 | SIFT-LLC | 2.4339 | 1.5626 | 1.1441 | 1.0061 |
3 | SIFT-FV |
\(\mathbf { 2.5470}\)
|
\(\mathbf { 2.3111} \)
|
\(\mathbf { 1.1484}\)
|
\(\mathbf { 1.0132} \)
|
4 | HOG-SPM | 1.6340 | 1.6010 | 1.0959 | 1.0042 |
5 | OB | 1.0003 | 1.0001 | 1.0001 | 1.0000 |
6 | SIFT-SPM | 1.4521 | 1.3573 | 1.0516 | 1.0052 |
7 | SURF-ScSPM | 1.8323 | 1.7330 | 1.0525 | 1.0060 |
8 | CENTRIST | 1.0045 | 1.0039 | 1.0000 | 1.0000 |
9 | GIST | 1.0604 | 1.0283 | 1.0047 | 1.0008 |
10 | LBP | 1.0434 | 1.0136 | 1.0021 | 1.0004 |
11 | Uniform LBP | 1.0187 | 1.0070 | 1.0015 | 1.0003 |
12 | LBP-HF | 1.0146 | 1.0091 | 1.0012 | 1.0002 |
13 | PLBP | 1.0630 | 1.0255 | 1.0034 | 1.0002 |
14 | HMAX | 1.0006 | 1.0003 | 1.0004 | 1.0001 |