摘要
人脸表情识别可看作是细粒度图像分类问题,由于表情细微的类间差异和显著的类内变化导致识别困难,为此设计了结合双线性卷积神经网络和注意力机制的人脸表情识别方法。该方法改进B-CNN算法,引入注意力模块,增强人脸表情的特征学习;采用基于Softmax损失函数与中心损失函数联合学习的策略,提高算法识别正确率。在CK+和RAF-DB公开数据集上算法识别率最高分别达到95.87%、90.15%。实验表明,所提方法能有效提高表情识别正确率,具有一定的实用价值。
Facial expression recognition could be regarded as a fine-grained image classification problem.Because of the subtle inter class differences and significant intra class changes of facial expression,it was difficult to recognize.Therefore,a facial expression recognition method combining bilinear convolution neural network and attention mechanism was designed.This method improved the B-CNN algorithm and introduced attention module to enhance the feature learning of facial expression;The strategy of joint learning based on softmax loss function and central loss function was a-dopted to improve the recognition accuracy of the algorithm.The recognition rate of the algorithm was the highest on CK+and RDF-DB public data sets,reaching 95.87%and 90.15%respectively.Experiments showed that the proposed method could effectively improve the accuracy of expression recognition,and had a certain practical value.
作者
徐迎春
XU Yingchun(Department of Intelligent Information,JiangHai Polytechnic College,Yangzhou,Jiangsu 225001,China)
出处
《九江学院学报（自然科学版）》
CAS
2022年第2期54-58,共5页
Journal of Jiujiang University：Natural Science Edition
基金
江海职业技术学院2021教学质量提升年资助课题(编号2021jhky003)的成果之一。
