Author Affiliations
School of Electrical & Control Engineering, Xi'an University of Science & Technology, Xi'an, Shaanxi 710054, Chinashow less
Fig. 1. Segmentation schematic of AR-LGC operator
Fig. 2. Coding process for improved asymmetric LGC
Fig. 3. Schematic of image fusion coding
Fig. 4. Extraction effects of various algorithms before and after adding Gabor wavelet. (a) Raw face image; (b) LBP operator before adding Gabor wavelet; (c) AR-LGC operator before adding Gabor wavelet; (d) multi-feature fusion coding before adding Gabor wavelet; (e) LBP operator after adding Gabor wavelet; (f) AR-LGC operator after adding Gabor wavelet; (g) proposed method
Fig. 5. Flow chart of face recognition
Fig. 6. Recognition rate curve under different number of blocks
Fig. 7. Treatment effect and process of proposed method
Fig. 8. Part of images from different face databases. (a) Yale face database; (b) ORL face database
Fig. 9. Part of images from different face databases. (a) FERET face database; (b) CMU-PIE face database
Number of blocks | Feature dimension | Average recognition rate /% |
---|
2×2 | 640 | 85.88 | 3×3 | 1440 | 88.80 | 4×4 | 2560 | 90.28 | 5×5 | 4000 | 95.03 | 9×9 | 12960 | 92.70 | 11×11 | 19360 | 89.30 | 16×16 | 40960 | 88.82 |
|
Table 1. Feature dimension and average recognition rate under different number of blocks
Method | Recognition rate |
---|
N=4 | N=5 | N=6 | N=7 | N=8 |
---|
PCA[10] | 76.4 | 79.3 | 87.6 | 91.5 | 93.2 | Gabor[10] | 84.1 | 90.0 | 92.0 | 92.3 | 94.1 | HOG informationentropy weighting[26] | 85.3 | 89.1 | 93.1 | 94.5 | 97.8 | Ref. [27] | 92.0 | 92.2 | 93.5 | 94.8 | 96.3 | LBP operator | 84.5 | 88.9 | 93.8 | 94.5 | 94.9 | AR-LGC operator | 85.0 | 92.8 | 94.6 | 95.0 | 95.1 | Proposed method | 85.5 | 93.2 | 96.0 | 97.2 | 97.9 |
|
Table 2. Recognition rate of each algorithm in YALE face database unit: %
Method | Recognition rate |
---|
N=4 | N=5 | N=6 | N=7 | N=8 |
---|
PCA[10] | 77.3 | 81.4 | 91.5 | 92.7 | 93.5 | Gabor[10] | 84.5 | 89.1 | 91.7 | 93.1 | 94.0 | HOG information entropy weighting[26] | 87.8 | 92.0 | 94.3 | 96.4 | 98.3 | Ref. [27] | 89.0 | 90.5 | 94.5 | 96.0 | 96.5 | LBP operator | 84.9 | 90.5 | 93.4 | 95.1 | 95.8 | AR-LGC operator | 85.2 | 92.6 | 94.5 | 95.8 | 96.0 | Proposed method | 85.4 | 93.9 | 96.1 | 97.6 | 98.5 |
|
Table 3. Recognition rate of each algorithm in ORL face database unit: %
Method | Recognition rate |
---|
be | bj | bf |
---|
LBP operator | 65.8 | 68.1 | 65.5 | AR-LGC operator | 83.9 | 89.5 | 81.6 | Proposed method | 89.4 | 93.8 | 87.7 |
|
Table 4. Recognition rate of each algorithm in FERET face database unit: %
Method | Recognition rate |
---|
N=10 | N=20 | N=30 | N=40 | N=50 |
---|
LBP operator[25] | 69.1 | 72.6 | 80.4 | 82.8 | 85.5 | AR-LGC operator | 71.0 | 75.2 | 85.1 | 86.0 | 88.4 | Ref. [28] | 73.8 | 83.4 | 86.2 | -- | -- | Ref. [25] | 74.8 | 84.7 | 87.6 | -- | -- | Ref. [29] | 78.6 | 87.0 | 90.0 | 90.2 | 90.3 | Proposed method | 75.8 | 80.6 | 90.2 | 91.3 | 94.6 |
|
Table 5. Recognition rate of each algorithm in CMU-PIE face database unit: %
Method | Average recognition time |
---|
LBP operator[25] | 22.36 | AR-LGC operator | 24.15 | Ref. [28] | 26.77 | Ref. [25] | 17.56 | Ref. [29] | 25.31 | Proposed method | 26.98 |
|
Table 6. Average recognition time of each algorithm in CMU-PIE face database unit: ms