Acta Optica Sinica, Volume. 38, Issue 9, 0910001(2018)
Supervised Sparsity Preserving Projection Based on Global Constraint
Figures & Tables(14)
Fig. 1. Sparsity reconstruction weights distribution of 50 samples by SPP algorithm on LFW database
Fig. 2. Sparsity reconstruction weights distribution of 50 samples by SSPP-GC algorithm on LFW database (a) with compactness constraint and (b) without compactness constraint
Fig. 3. 2D visualizations on Extended Yale B database. (a) 2D visualizations of SPP; (b) 2D visualizations of SSPP-GC
Fig. 4. Partial sample images on four databases. (a) Partial samples of AR database; (b) partial samples of Extended Yale B database; (c) partial samples of LFW database; (d) partial samples of PubFig database
Fig. 5. Comparison of sub-space recognition results of different algorithms in different dimensions. (a) AR database; (b) Extended Yale B database
Fig. 6. Samples of an individual from Extended Yale B database. (a) Add noise images; (b) add occlusion images
Table 1. Performance comparisons of different algorithms and their corresponding optimal dimensions on four databases
View tableTable 1. Performance comparisons of different algorithms and their corresponding optimal dimensions on four databases
Algorithm Optimal recognition rate Sub-space dimension AR Extended Yale B LFW PubFig AR Extended Yale B LFW PubFig PCA 85.43 91.60 35.79 32.12 213 170 322 290 LDA 89.14 92.44 55.89 26.16 99 37 61 98 LPP 89.57 94.07 56.30 25.76 91 115 65 108 NPE 90.14 95.59 55.10 25.05 191 151 181 271 LSDA 90.57 92.68 56.30 27.27 213 151 151 251 SPP 91.00 92.62 41.45 30.51 213 151 322 281 SSPP-GC 93.86 95.04 62.01 36.16 185 101 59 245
Table 2. Performance comparisons of different algorithms and their corresponding optimal dimensions on the AR and Extended Yale B databases
View tableTable 2. Performance comparisons of different algorithms and their corresponding optimal dimensions on the AR and Extended Yale B databases
Algorithm Optimal recognition rate Sub-space dimension AR Extended Yale B AR Extended Yale B DSNPE[ 26 ]91.43 94.62 150 140 SSPP-GC 93.86 95.04 185 101 WDSPE[ 27 ]82.50 87.89 180 200 SSPP-GC 91.71 89.28 120 80 GRSDA[ 28 ]93.8 93.4 130 361 SSPP-GC 98.00 98.19 60 110
Table 3. Performance comparison of different algorithms on the LFW and PubFig databases
View tableTable 3. Performance comparison of different algorithms on the LFW and PubFig databases
Database Method DSNPE[ 26 ]SVDL[ 38 ]FDDL[ 19 ]SSPP-GC LFW 56.26 57.47 60.84 62.01 PubFig 35.35 37.45 27.27 36.16
Table 4. Sparse reconstruction time and low dimensional projection time of SSPP-GC and SPP on four databasess
View tableTable 4. Sparse reconstruction time and low dimensional projection time of SSPP-GC and SPP on four databasess
Time AR Extended Yale B LFW PubFig SPP SSPP-GC SPP SSPP-GC SPP SSPP-GC SPP SSPP-GC t C3628.21 25.59 4455.12 79.45 2810.23 44.23 10628.27 63.74 t P0.21 0.35 0.48 0.24 0.33 0.51 0.69 0.56
Table 5. Recognition rates of different algorithms with different classifiers on the LFW database%
View tableTable 5. Recognition rates of different algorithms with different classifiers on the LFW database%
Classifier Method PCA LDA LPP NPE LSDA SPP SSPP-GC KNN 15.02 38.45 39.70 32.67 23.55 33.50 62.80 SVM 16.27 48.15 49.23 57.01 53.27 54.14 62.01 SRC 35.79 55.89 56.30 55.10 56.30 41.45 62.01
Table 6. Recognition rates of different algorithms with different classifiers on the PubFig database%
View tableTable 6. Recognition rates of different algorithms with different classifiers on the PubFig database%
Classifier Method PCA LDA LPP NPE LSDA SPP SSPP-GC KNN 14.95 12.53 11.62 6.06 5.96 27.07 33.23 SVM 12.53 18.28 19.19 21.11 22.22 31.82 39.09 SRC 32.12 26.16 25.76 25.05 27.27 30.51 36.16
Table 7. Comparison of recognition rates of di?erent occlusions on Extended Yale B database
View tableTable 7. Comparison of recognition rates of di?erent occlusions on Extended Yale B database
Algorithm Recognition rates /% Optimal projection dimension 2 pixel× 2 pixel 4 pixel× 4 pixel 8 pixel× 8 pixel 16 pixel× 16 pixel 2 pixel× 2 pixel 4 pixel× 4 pixel 8 pixel× 8 pixel 16 pixel× 16 pixel PCA 90.48 90.48 89.98 87.65 159 159 158 142 LDA 91.25 91.14 90.20 88.04 37 37 37 37 LPP 92.58 92.58 92.36 89.20 101 101 101 99 NPE 94.68 94.46 94.24 91.92 159 149 14 132 LSDA 91.97 91.97 91.03 88.59 159 159 158 142 SPP 92.97 92.91 92.36 90.20 149 159 158 142 SSPP-GC 95.04 95.02 94.13 92.91 101 101 81 81
Tools
Get Citation
Copy Citation Text
Ying Tong, Yimin Wei, Yuehong Shen. Supervised Sparsity Preserving Projection Based on Global Constraint[J]. Acta Optica Sinica, 2018, 38(9): 0910001
Paper Information
Category: Image Processing
Received: Jan. 8, 2018
Accepted: Apr. 16, 2018
Published Online: May. 9, 2019
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20