2021 IEEE International Conference on Image Processing
19-22 September 2021 • Anchorage, Alaska, USA
Imaging Without Borders
| Paper ID | ARS-3.8 | ||
| Paper Title | DFER-NET: RECOGNIZING FACIAL EXPRESSION IN THE WILD | ||
| Authors | Yumin Tian, Mengqi Li, Di Wang, Xidian University, China | ||
| Session | ARS-3: Image and Video Biometric Analysis | ||
| Location | Area H | ||
| Session Time: | Monday, 20 September, 13:30 - 15:00 | ||
| Presentation Time: | Monday, 20 September, 13:30 - 15:00 | ||
| Presentation | Poster | ||
| Topic | Image and Video Analysis, Synthesis, and Retrieval: Image & Video Interpretation and Understanding | ||
| IEEE Xplore Open Preview | Click here to view in IEEE Xplore | ||
| Abstract | Recently, deep convolutional neural networks have made remarkable progress in facial expression recognition. However, they often fail in the natural environment, which is due to two challenging issues. One is that facial expression data in real world often has imbalanced distribution. The other is the intra- and inter- variations caused by changes in head pose, illumination and occlusions. In this paper, a discriminative facial expression recognition network (DFER-Net) is proposed to recognize facial expression in the wild by maximizing the intra-class similarity while minimizing inter-class similarity as well as strengthening the weight of minority class. Specifically, DFER-Net adds two fully-connected layers of a novel quadruplet-mean loss and a decision layer of a novel balanced-softmax loss on the traditional deep convolutional neural network. The quadruplet-mean loss enlarges intra-class similarity and inter-class distinction with high efficiency, which enhances the discriminative power of the DFER-Net. And the balanced-softmax loss strengthens the weight of minority class, which solves the class imbalance problem. Extensive experiments on benchmark datasets show the superior performance of the proposed DFER-Net over baseline methods in facial expression recognition. | ||