Facial Expression Recognition Based on Key Region Feature Extraction
Nengsheng Bao, Qingjiang Xiang, Yuping Yang, Tianyu Huang · 2025
Although deep learning techniques have achieved significant success in image-based expression recognition, they still face challenges such as insufficient key region feature capture, difficulty in distinguishing difficult samples, dealing with class imbalance issues, and improving model generalization capability. To address these problems, we propose a network based on Key Region Feature Extraction. Our method extracts focus region features and global region features in face images by extracting them separately, and then fuses the extracted local region features and global features at different spatial scales using a multi-scale region fusion mechanism, so that the network can capture the global and local features and the relationship between the features in the image more efficiently and achieve a better discriminative performance among different expression categories. Additionally, we use a loss function “focal loss“ for the problem of category imbalance and difficult samples in model training to improve the model's ability to recognize a few categories. Finally, our method achieves n average accuracy of 79.982% on the FER2013, 92.14% on the RAFDB, and 68% on the Affect-Net.