Table 7.
Performance comparison of the proposed cascaded CNN structure with different studies
| Methods | Database | #Expression | Accuracy (%) |
|---|---|---|---|
| HoG + NNE [6] | RaFD, TFEID, JAFFE | 5 | 93.75 |
| Facial Components Detection + KNN [36] | RaFD | 6 | 75.61 |
| Viola & Jones + AAM +ANN [13] | RaFD | 7 | 89.55 |
| Surf Boosting [61] | RaFD | 6 | 90.64 |
| Facial Components Detection + Fuzzy [36] | RaFD | 6 | 93.96 |
| CNN [26] | RaFD | 6 | 94.16 |
| Cascade CNN [81] | RaFD | 6 | 93.43 |
| LBP + SVM [2] | MUG | 7 | 77.14 |
| LBP + Geometric Features + SVM [28] | MUG | 6 | 83.12 |
| CNN [26] | MUG | 6 | 87.68 |
| Gabor + NN [19] | MUG | 6 | 89.29 |
| PCA + SRC [5] | MUG | 7 | 91.27 |
| Landmark points + SVM [3] | MUG | 6 | 92.76 |
| Proposed method | RaFD(3-channel raw image) | 6 | 89.44 |
| RaFD (1-channel final-iconize) | 6 | 90.55 | |
| MUG (4-channel combine) | 6 | 93.33 | |
| RaFD (4-channel combine) | 6 | 94.44 |