Fig. 1. Flow chart of our algorithm
Fig. 2. Diagram of direction vector decomposition
Fig. 3. Point extracted in the nose root. (a) Points extracted along the direction vector ss-o; (b) point searched in the nose
Fig. 4. Extracted areas of the nose and eyes. (a) Nose area; (b) eyes area; (c) areas combing nose with eyes
Fig. 5. Definition process of coverage rate. (a) Feature points location; (b) feature points selection; (c) range partition; (d) coverage area
Fig. 6. Combination areas with different coverage rates. (a) 25% coverage rate; (b) 50% coverage rate; (c) 75% coverage rate; (d) 90% overage rate
Fig. 7. Combination area under different coverage rates after keypoints filtering. (a) 25% coverage rate; (b) 50% coverage rate; (c)75% coverage rate; (d) 90% coverage rate
Fig. 8. Four isogeodesic curves extracted in this paper
Coverage rate | Accuracy |
---|
Bosphorus | FRGCv2.0 |
---|
25 | 92.98 | 92.66 | 50 | 93.62 | 94.87 | 75 | 95.54 | 97.16 | 90 | 94.12 | 96.59 |
|
Table 1. Face recognition rates in terms of different coverage rates unit: %
Fusion coefficient μ | Accuracy |
---|
Bosphorus | FRGCv2.0 |
---|
0.1 | 89.12 | 89.63 | 0.2 | 90.56 | 91.96 | 0.3 | 92.66 | 93.23 | 0.4 | 93.98 | 95.66 | 0.5 | 95.93 | 97.98 | 0.6 | 97.01 | 98.63 | 0.7 | 96.43 | 98.23 | 0.8 | 96.03 | 97.98 | 0.9 | 95.95 | 97.25 |
|
Table 2. Face recognition rates in terms of different fusion coefficients unit:%
Algorithm | Neutral | Anger | Disgust | Fear | Happy | Sadness | Surprise | All |
---|
In Ref. [19] | 100.00 | 95.80 | 91.30 | 95.70 | 96.20 | 97.00 | 95.80 | 96.80 | In Ref. [20] | 100.00 | 95.80 | 88.40 | 81.40 | 88.70 | 97.00 | 93.00 | 93.40 | In Ref. [7] | 98.96 | 94.12 | 88.24 | 98.55 | 98.08 | 96.92 | 100.00 | 97.41 | Ours | 100.00 | 95.71 | 92.11 | 96.82 | 97.54 | 97.66 | 97.13 | 97.01 |
|
Table 3. Comparison of different algorithms on the Bosphorus database unit:%
Algorithm | Neutral | Non-neutral |
---|
DA-NICP | 100.00 | 97.00 | In Ref. [6] | 100.00 | 98.52 | In Ref. [7] | 98.45 | 97.50 | In Ref. [19] | 100.00 | 96.30 | CPD | 100.00 | 97.91 | In Ref. [9] | 98.88 | 98.85 | In Ref. [10] | 98.70 | 98.70 | Ours | 100.00 | 98.63 |
|
Table 4. Comparison of different algorithms on the FRGCv2.0 database unit:%