基于PCA与深度学习的表情识别算法  

FACIAL EXPRESSION RECOGNITION BASED ON PCA AND DEEP LEARNING

作　　者：区健 冯开平 罗立宏 Ou Jian;Feng Kaiping;Luo Lihong(School of Computer Science,Guangdong University of Technology,Guangzhou 510006,Guangdong,China;School of Art and Design,Guangdong University of Technology,Guangzhou 510090,Guangdong,China)

机构地区：[1]广东工业大学计算机学院,广东广州510006 [2]广东工业大学艺术与设计学院,广东广州510090

出　　处：《计算机应用与软件》2023年第7期185-191,共7页Computer Applications and Software

基　　金：教育部人文社科项目(20YJAZH073)。

摘　　要：针对传统特征提取主观性强、识别率不高,而现有卷积神经网络层数较多、规模过大、结构不够轻量化等问题,提出一种结合主成分分析(PCA)和卷积神经网络的人脸表情识别方法。使用主成分分析对数据集原始图片进行降维重构,去除与面部表情无关的冗余信息,将图像进行数据增强之后作为输入图像输入到改进神经网络模型进行识别。该网络在输入层加入一层非线性表示层,并在全连接层前采用全局平均池化保留局部信息,结构较为轻量,参数量少。在公开的数据集FER2013、JAFFE中进行实验,并且对实验结果进行分析。实验结果表明,该方法在FER2013和JAFFE数据集的准确率达到了69.07%和98.23%。所提方法有效地提高人脸识别的性能,为人脸表情识别领域改进识别率方向提供了一个思路。Aimed at the problems of traditional feature extraction with strong subjectivity,low recognition rate,and the existing convolutional neural networks with many layers,too large scale,and not enough lightweight structure,this paper proposes a facial expression recognition method combining principal component analysis(PCA)and convolutional neural network.PCA was used to perform dimensionality reduction and reconstruction on the original images of the data set,so as to remove the redundant information unrelated to facial expressions.The image after data augmentation was input as an input image to the improved neural network model for recognition.The improved network added a non-linear representation layer to the input layer,and used global average pooling to retain local information before the fully connected layer.The structure was relatively lightweight and the number of parameters was small.Experiments were conducted on the public datasets FER2013 and JAFFE,and the experimental results were analyzed.Experimental results show that the accuracy of the proposed method on the FER2013 and JAFFE datasets reaches 69.07%and 98.23%.The proposed method effectively improves the performance of face recognition and provides a way to improve the recognition rate in the field of facial expression recognition.

关 键 词：计算机图像处理 人脸表情识别 主成分分析 深度学习 卷积神经网络 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]