| 摘 要: 针对常用人脸表情识别算法在无标签跨域情况下难以取得好的识别效果的挑战,提出一种基于子空间学习的跨域人脸表情识别算法。该算法首先构建边缘分布和条件分布双对齐约束来减小域分布差异;其次构造稀疏数据矩阵;最后联合双分布对齐约束和结构保持约束训练得到重构子空间,通过学习得到跨域人脸表情不变特征表达,输入支持向量机(SVM)分类器完成跨域人脸表情识别。实验结果表明,本算法取得了59.97%的平均识别率,相对次优方法效果提升了2.02%,验证了本算法的有效性。 |
| 关键词: 对齐约束 结构保持约束 联合约束 重构子空间 跨域人脸表情识别 |
|
中图分类号: TP391.41
文献标识码: A
|
| 基金项目: 安徽省高校自然科学研究重点项目(KJ2021A1255);安徽省优秀青年教师培养项目(YQYB2023070);安徽开放大学老年教育重点课题(LNJY2022ZD06) |
|
| Cross-Domain Facial Expression Recognition Based on Jointly Constrained Reconstructed Subspace Learning |
|
QI Mei
|
(School of Information and Construction Engineering, Anhui Open University, Hefei 230022, China)
qimei620@163.com
|
| Abstract: In response to the challenge of achieving satisfactory recognition performance with common facial expression recognition algorithms under unlabeled cross-domain conditions, a cross-domain facial expression recognition method based on subspace learning is proposed. This method first constructs dual alignment constraints for both marginal and conditional distributions to reduce domain distribution discrepancies. Subsequently, a sparse data matrix is constructed. Finally, by jointly integrating the dual distribution alignment constraints and structural preservation constraints, a reconstructed subspace is trained to learn invariant cross-domain facial expression feature representations, which are then input into an SVM(Support Vector Machine)classifier to accomplish cross-domain facial expression recognition. Experiment results demonstrate that the proposed method achieves an average recognition rate of 59.97% , outperforming the second-best method by 2.02% , thereby validating the effectiveness of the algorithm. |
| Keywords: alignment constraint structural preservation constraint joint constraint reconstructed subspace cross-domain facial expression recognition |
