Table 1.
Comparison between FER approaches.
| Approach | Method | Human Resource | Computing Resource | Computational Complexity | Accuracy |
|---|---|---|---|---|---|
| Conventional | Gabor wavelets coefficients [7] | High | Low | Low | Medium |
| Haar features [8] | High | Low | Low | Medium | |
| Local binary pattern (LBP) [9] | High | Low | Low | Medium | |
| LBP on three orthogonal planes (LBP-TOP) [10] | High | Low | Low | Medium | |
| Scale-invariant feature transform (SIFT) [29] | High | Low | Low | Medium | |
| Histogram of gradient (HOG) [11] | High | Low | Low | Medium | |
| Deep learning-based | Convolutional neural network [40] | Low | High | High | High |
| Transfer learning-based CNN (Our approaches) | Low | Medium | Medium | High |