摘要
针对图像表情判别精度低下的问题,提出一种基于改进型VGG-16网络的人脸表情识别方法。为解决传统方法存在像素特征分布不均的问题,采用基于改进的高斯混合模型进行图像特征数据的有效提取;基于改进的VGG-16深度神经网络,增强人脸表情识别的训练样本,实现对采集的图像数据多表情多场景精准区分。基于通用数据集及自采集数据集进行仿真实验,验证所提方法在表情识别的准确度和速度方面都展现出一定优势,尤其在黑暗条件下识别准确率可达90%左右。
Aiming at the problem of low accuracy of image expression discrimination,a facial expression recognition method based on improved VGG-16 network was proposed.To solve the problem of uneven pixel feature distribution in traditional methods,an improved Gaussian mixture model was used to effectively extract image feature data.Based on the improved VGG-16 deep neural network,the training samples for facial expression recognition were enhanced to achieve accurate discrimination of the collected image data with multiple expressions and multiple scenes.Simulation experiments based on a common data set and self-collected data set,verify that the proposed method in terms of accuracy and speed expression recognition shows some advantages,especially in dark conditions,its recognition accuracy rate reaches 90%.
作者
程学军
邢萧飞
CHENG Xue-jun;XING Xiao-fei(College of Information Engineering,Luohe Institute of Technology,Henan University of Technology,Luohe 462000,China;School of Computer Science and Cyber Engineering,Guangzhou University,Guangzhou 528225,China)
出处
《计算机工程与设计》
北大核心
2022年第4期1134-1144,共11页
Computer Engineering and Design
基金
河南省科技攻关计划基金项目(222102110011)
国家自然科学基金河南省联合基金项目(U1604149)
河南省教育厅自然科学基金项目(19A520006)。
关键词
表情识别
VGG-16网络模型
高斯混合模型
相关情绪标签分布学习
正则化学习
红外图像
expression recognition
VGG-16 network model
Gaussian mixture model
correlation emotion label distribution learning
regularization learning
infrared image
