Author Affiliations
1 School of Physics and Optoelectronics, Xiangtan University, Xiangtan, Hunan 411105, China2 College of Mechanical and Electrical Engineering, Hunan Applied Technology University,Changde, Hunan 415000, Chinashow less
Fig. 1. Sobel operator. (a) Horizontal direction; (b) vertical direction; (c) 45° direction; (d) 135° direction
Fig. 2. (a) Sampling point; (b) LBP encoding; (c) GLBP encoding
Fig. 3. Robustness of GLBP versus LBP. (a) Original encoding; (b) noise encoding
Fig. 4. (a) Traditional feature extraction; (b) improved feature extraction
Fig. 5. IGLBP encoding process
Fig. 6. Flow chart of the proposed algorithm
Fig. 7. Flow chart of IGLBP feature extraction
Fig. 8. (a) Training sample; (b) AR expression subset; (c) AR illumination subset; (d) AR partial occlusion subset A; (e) AR partial occlusion subset B
Fig. 9. (a) Training sample; (b) CAS-PEAL expression subset; (c) CAS-PEAL background subset; (d) CAS-PEAL accessory subset
Fig. 10. YALE face library
Algorithm | Facialexpressionsubset | Illuminationsubset | Partialocclusionsubset A | Partialocclusionsubset B |
---|
LBP | 93.33 | 92.33 | 91.67 | 72.67 | LDP | 96.33 | 93.00 | 90.00 | 71.67 | CSLBP | 96.67 | 95.00 | 87.67 | 73.00 | LGBP | 94.67 | 99.00 | 94.33 | 90.33 | IGLBP | 99.67 | 99.00 | 98.67 | 94.33 |
|
Table 1. Recognition rate on AR database of different algorithms%
Algorithm | Backgroundsubset | Expressionsubset | Accessorysubset |
---|
LBP | 91.00 | 89.00 | 91.00 | LDP | 97.33 | 95.00 | 86.00 | CSLBP | 93.25 | 87.25 | 90.25 | LGBP | 97.52 | 91.25 | 93.00 | IGLBP | 99.75 | 96.25 | 97.00 |
|
Table 2. Recognition rate on CAS-PEAL database of different algorithms%
Algorithm | Number of sample |
---|
2 | 3 | 4 | 5 |
---|
LBP | 73.25 | 77.66 | 78.00 | 78.33 | LDP | 78.00 | 81.75 | 82.34 | 84.75 | CSLBP | 79.37 | 82.06 | 83.00 | 85.72 | LGBP | 85.32 | 90.09 | 92.33 | 93.25 | IGLBP | 86.61 | 92.05 | 93.57 | 94.53 |
|
Table 3. Recognition rate on YALE database of different algorithms%
Algorithm | σ=0 | σ=0.0001 | σ=0.0002 | σ=0.0003 | σ=0.0004 | δ |
---|
LBP | 91.00 | 36.25 | 16.00 | 9.25 | 6.75 | 92.58 | LDP | 97.33 | 95.00 | 93.25 | 90.50 | 87.75 | 9.84 | CSLBP | 93.25 | 90.25 | 83.52 | 76.25 | 65.25 | 30.03 | LGBP | 97.52 | 96.25 | 95.75 | 94.00 | 93.25 | 4.38 | IGLBP | 99.75 | 99.50 | 99.25 | 99.00 | 98.50 | 1.25 |
|
Table 4. Results on CAS-PEAL background subset of different algorithms after adding different variances of Gaussian noise%
Algorithm | σ=0 | σ=0.0001 | σ=0.0002 | σ=0.0003 | σ=0.0004 | δ |
---|
LBP | 89.00 | 28.50 | 13.50 | 7.25 | 6.25 | 92.98 | LDP | 95.00 | 91.25 | 87.25 | 82.98 | 78.00 | 17.89 | CSLBP | 87.25 | 83.50 | 79.25 | 74.75 | 63.25 | 27.51 | LGBP | 91.25 | 90.50 | 89.00 | 88.50 | 86.75 | 4.93 | IGLBP | 96.25 | 92.50 | 92.00 | 91.25 | 90.75 | 5.71 |
|
Table 5. Results on CAS-PEAL expression subset of different algorithms after adding different variances of Gaussian noise%
Algorithm | σ=0 | σ=0.0001 | σ=0.0002 | σ=0.0003 | σ=0.0004 | δ |
---|
LBP | 91.00 | 26.25 | 14.00 | 8.50 | 6.50 | 92.86 | LDP | 86.00 | 83.50 | 79.25 | 74.75 | 71.67 | 16.66 | CSLBP | 90.25 | 97.25 | 81.00 | 74.75 | 63.00 | 30.19 | LGBP | 93.00 | 91.25 | 90.50 | 89.75 | 87.65 | 5.75 | IGLBP | 97.00 | 94.75 | 94.50 | 93.50 | 92.75 | 4.90 |
|
Table 6. Results on CAS-PEAL accessory subset of different algorithms after adding different variances of Gaussian noise%
Algorithm | Feature dimension | T1 /ms | T2 /ms |
---|
LBP | 16384 | 30.3 | 23.2 | LDP | 16384 | 439.1 | 22.9 | CSLBP | 1024 | 23.6 | 2.5 | LGBP | 655360 | 883.5 | 102.6 | IGLBP | 16384 | 94.5 | 23.0 |
|
Table 7. Feature dimensions and average time of different algorithms on YALE database