基于面部特征检测的人脸表情实时识别

苏萃文 柴国强

doi:10.15966/j.cnki.dnydx.2023.z1.018

电脑与电信 >

2023 , Vol. 1 >Issue 1-2: 17 - 21

DOI: https://doi.org/10.15966/j.cnki.dnydx.2023.z1.018

基金项目

基于面部特征检测的人脸表情实时识别

展开

山西师范大学物理与信息工程学院

网络出版日期: 2023-05-29

基金资助

山西省基础研究计划,项目编号：20210302124647；山西省高等学校科技创新项目,项目编号：2021L269；山西省高等学校大学生创新创业训练计划项目,项目编号：20220310；山西师范大学大学生创新创业训练计划资助项目,项目编号：2022DCXM-52；

收起

Real-time Facial Expression Recognition Based on Facial Feature Detection

Expand

Shanxi Normal University

Online published: 2023-05-29

Fold

摘要

针对传统方法特征提取识别准确率低、提取方法复杂以及深度学习方法模型复杂、实时性较差等缺点，提出了一种基于OpenCV和Dlib模型的人脸表情实时识别方法。首先，使用OpenCV实时采集并预处理图像；然后，使用Dlib预训练模型提取人脸特征点；最后，结合本文提出的5个表情识别指标：眉毛的倾斜程度、眼睛的睁开程度、上嘴唇与鼻尖高度占比、嘴巴宽度占比和嘴巴高度占比，对人脸表情进行实时识别。通过搭建系统界面，简化操作，增强了识别方法的实用性。实验表明，所提出算法表情识别的准确率均到达96%以上，验证了其有效性。

关键词： ; uniplatform=NZKPT" name="keyword">深度学习; ; uniplatform=NZKPT" name="keyword">人脸表情实时识别; ; uniplatform=NZKPT" name="keyword">OpenCV; ; uniplatform=NZKPT" name="keyword">Dlib模型;

本文引用格式

苏萃文柴国强 .

基于面部特征检测的人脸表情实时识别

[J]. 电脑与电信, 2023 , 1(1-2) : 17 -21 . DOI: 10.15966/j.cnki.dnydx.2023.z1.018

Abstract

A real-time facial expression recognition based on OpenCV and Dlib model is proposed in this paper, which

avoids low recognition rate and complex extraction process in traditional methods and complex model and poor real-time

in deep learning-based methods. First, OpenCV is used to capture and pre-process images in real time. Then the Dlib

model is applied to calibrate the facial key points in the acquired face image. Finally, fifive indexes proposed in the paper,

i.e., eyebrow tilt degree, eyes open degree, upper lip to nasal tip height ratio, mouth width ratio and mouth height ratio are

combined to recognize expression. Besides, a system interface with simplifified operation is built to enhance the

practicability of the proposed method. Experiments show that the accuracy of the proposed method is above 96%, which

verififies its effffectiveness.

Key words：

deep learning

; real-time facial expression recognition; OpenCV; Dlib

Options

摘要页面

文章导航

模态框（Modal）标题

摘要

本文引用格式

Abstract