Table 1.
Summary of the state-of-the-art for automated pain level detection using facial images.
| Study | Method | Classifier | Number of subjects | Number of facial images/frames | Accuracy (%) | Limitations |
|---|---|---|---|---|---|---|
| Brahnam et al.11 | Principal component analysis, linear discriminant analysis | Support vector machine | 26 (13 male, 13 female) | 204 | 88.00 | Single and small dataset, low accuracy |
| Brahnam et al.12 | Principal component analysis, linear discriminant analysis, frequency domain methods | Neural network simultaneous optimization algorithm | 26 (13 male, 13 female) | 204 | 100.0 | Single and small dataset |
| Brahnam et al.13 | Principal component analysis, linear discriminant analysis | Neural network simultaneous optimization algorithm | 26 (13 male, 13 female) | 204 | 90.20 | Single and small dataset |
| Kristian et al.14 | Active shape model, local binary pattern | Support vector machine | 23 | 132 | 88.70 | Single and small dataset |
| Othman et al.16 | MobileNetV2 | Softmax | 1. 87 | 1. 3480 | 1. 6550 | Low accuracy |
| 2. 134 | 2. 7763 | 2. 7140 | ||||
| Weitz et al.19 | Convolutional neural network | Softmax | 324 | 14,322 | 67.00 | Single dataset, low accuracy |
| Yang et al.20 | Local binary pattern, local phase quantization, statistical features | Support vector machine | 1. 129 | 1. 48,398 | 1. 8342 | Low accuracy |
| 2. 90 | 2. 8700 | 2. 71.00 | ||||
| Kharghanian et al.22 | Convolutional deep belief network model | Support vector machine | 25 | 48,398 | 87.20 | Single dataset, low accuracy |
| Zafar and Khan23 | Geometric features | k-nearest neighbor | Unspecified | 21,500 | 84.02 | Single dataset, low accuracy |