Table 20 Unsolved challenges and a proposed design of solution.
From: Improved facial emotion recognition model based on a novel deep convolutional structure
Unsolved challenges | Solution in the proposed model | |
|---|---|---|
1 | Limitation no. of samples in CK+, Jaffee datasets. | Augmentation to increase no. of samples for high accurate training. |
2 | Unbalance samples in emotion folders in RAF dataset. | Augmentation to balance no. of samples in each emotion for efficient and accurate classification. |
3 | Poor quality and different real-world scenarios (occlusion) existing in RAF dataset | Proposing AA-DCN model to enhance the classification task of RAF, a real word dataset, (one of the most challenging dataset). |
4 | The anti-aliasing problem that appears due to down sampling in traditional CNN models, that leads to miss classification tasks | With challenging datasets like RAF, the suggested AA-DCN model successfully overcomes the anti-aliasing phenomena leading to a significant increase in FER task. |
