基于特征融合与选择的小样本表情识别  

Expression recognition of small samples based on feature fusion and selection

作　　者：王从澳 黄润才[1] 孙延标 杨彬 孙刘成 WANG Congao;HUANG Runcai;SUN Yanbiao;YANG Bin;SUN Liucheng(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》2022年第10期112-116,共5页Transducer and Microsystem Technologies

摘　　要：针对小样本数据集的数据量不足和深度学习的模型复杂、参数冗余等问题,提出了一种基于特征融合与选择的小样本表情识别算法。该模型将手工提取的人脸关键区域特征与DenseNet网络提取的深度学习特征相融合,再利用基于熵的特征约简技术对特征维度进行熵减和选择,以使用多分类支持向量机(MCSVM)进行识别分类,通过在JAFFE和CK+公开数据集上的实验测试结果表明:该模型在小样本数据集上具有更高的识别准确率和实时性,显著提升了人脸表情的识别性能。Aiming at the problems of insufficient data quantity, complex deep learning model and redundant parameters of small sample dataset, a small sample expression recognition algorithm based on feature fusion and selection is proposed.In this model, the key region features extracted by hand are combined with the deep learning features extracted by DenseNet network, and then the feature dimensions are reduced and selected by using entropy based feature reduction technology, so as to use multi-classification support vector machine(MCSVM)for classification and recognition.The experimental results on JAFFE and CK+ public datasets show that the model is effective in small sample datasets.It has higher recognition accuracy and real-time performance, and greatly improves the recognition performance of facial expression.

关 键 词：特征融合 DenseNet网络 特征约简 多分类支持向量机 表情分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]