摘要
由于单一的神经网络通道进行人脸表情识别会造成特征信息丢失,设计了一种特征融合的算法。首先将直方图均衡化处理后的面部表情图片通过卷积神经网络提取全局特征,将原图通过局部二值模式处理后,送入另一个卷积神经网络模型提取图片局部特征,再将全局特征图与局部特征图后加权融合后,通过softmax进行分类。在FER+数据集上测试得到了较好的识别结果,验证了模型的有效性。
Because a single neural network channel for facial expression recognition will cause the loss of feature information,an algorithm of feature fusion is designed in this paper.The histogram equalization processed facial expression pictures extract the global features through convolutional neural network,after processing the original map through the local two-value pattern,into another convolutional neural network model to extract the picture local features,and then the global feature map and local feature map after weighted fusion,through softmax classification.The results of the test on the FERplus data set are obtained and the validity of the model is verified.
出处
《工业控制计算机》
2020年第11期80-83,共4页
Industrial Control Computer
关键词
局部二值模式
表情识别
卷积神经网络
特征融合
local binary mode
expression recognition
convolutional neural network
feature fusion
