基于图像融合与深度学习的人脸表情识别  

Facial expression recognition based on image fusion and deep learning

作　　者：焦阳阳 黄润才[1] 万文桐 张雨 JIAO Yangyang;HUANG Runcai;WAN Wentong;ZHANG Yu(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201600,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201600

出　　处：《传感器与微系统》2024年第3期148-151,共4页Transducer and Microsystem Technologies

摘　　要：针对纹理特征提取方法单一及深度学习不能有效提取图像局部特征的问题,提出一种基于图像融合与深度学习的人脸表情识别方法。首先,对人脸表情图像分别提取局部二值模式(LBP)图像与韦伯局部描述符(WLD)图像;然后,将2种纹理图像进行融合作为输入图像送入改进后的残差神经网络(Res-Net)提取表情特征;将ResNet中的卷积核替换为空洞卷积,并在网络中添加改进后的注意力机制,使模型更加关注有效特征;最后,使用SoftMax进行表情分类。在JAFFE和CK+数据集上进行实验,准确率分别为97.0%与99.3%。实验结果表明,该方法能有效提高人脸表情识别的准确率。Aiming at the problem that the texture feature extraction method is single and deep learning cannot effectively extract image local features,a facial expression recognition method based on image fusion and deep learning is proposed.Firstly,the local binary pattern(LBP)image and the Weber local descriptor(WLD)image are extracted from the facial expression image.Then,the two texture images are fused as the input image and sent to the improved residual neural network(ResNet)to extract facial expression features,the convolution kernel in ResNet is replaced with dilated convolution,and an improved attention mechanism is added to the network to make the model pay more attention to effective features.Finally,SoftMax is used for expression classification.Experiments are performed on the JAFFE and CK+datasets,and the accuracy rates are 97.0%and 99.3%,respectively.The experimental results show that this method can effectively improve the accuracy of facial expression recognition.

关 键 词：人脸表情识别 注意力机制 卷积神经网络 特征提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]