摘要
针对传统机器学习方法在人脸表情识别上存在特征提取繁琐、表情识别准确率不高的问题,提出一种基于深度学习的人脸表情自动识别方法.设计了一个卷积神经网络模型,以原始图像数据为输入,中间以卷积层和池化层交替作为隐层进行特征自动提取,最后将提取到的特征数据映射到全连接层,并采用Softmax函数作为分类器计算分类得分概率,实现人脸表情的自动识别分类.在公开的人脸表情数据集CK+上进行实验,结果表明本文方法能更准确地识别人脸表情.
In order to solve the problems of feature extraction and low accuracy of facial expression recognition in traditional machine learning methods,an automatic facial expression recognition method based on deep learning is proposed.We designed a convolutional neural network model,which took the original image data as input and the intermediate convolutional layer and pooling layer were alternately used as hidden layer for feature automatic extraction.Finally,the extracted feature data was mapped to the full connection layer and Softmax function was used as the classifier to calculate the classification score probability in order to realize the automatic recognition and classification of facial expressions.Experiments were carried out on the public face expression data set CK+,and the results show that this method can recognize face expression more accurately.
作者
邹建成
邓豪
ZOU Jiancheng;DENG Hao(Institute of Image Processing and Pattern Recognition,North China Univ.of Tech.,100144,Beijing,China)
出处
《北方工业大学学报》
2019年第5期51-56,共6页
Journal of North China University of Technology
基金
国家自然科学基金项目“微阵列相机动态场景的超分辨率和测量研究”(61572038)
关键词
表情识别
特征提取
深度学习
卷积神经网络
facial recognition
feature extraction
deep learning
convolutional neural network
