摘要
在人脸识别方向,文章提出了一种基于稠密卷积神经网络的人脸表情识别方法。该方法通过特征复用与旁路连接的策略,实现低计算资源的高效特征表达,从而达到提高表情识别系统准确率的目的。文中提出的人脸表情识别方法,使用GPU运算优化模型,最终在拓展的KDEF人脸表情数据集上取得了96.88%的准确率,优于目前已公开的大部分人脸表情识别方法。文章利用预先训练好的人脸表情识别模型,设计出表情识别原型软件,通过使用摄像头实时捕捉人脸表情进行识别,能够取得很好的识别效果。
In the face recognition field,this paper proposes a human facial expression recognition method based on dense convolutional neural network.The method achieves efficient feature expression with low computing resources through feature reuse and bypass connection strategy,so as to improve the accuracy of facial expression recognition system.The human facial expression recognition method proposed in this paper uses GPU computing to continuously research and optimize the model,and finally it achieves 96.88%accuracy on the human facial expression database which is composed with KDEF dataset and the human facial expression dataset produced in the article,which is better than the most of the current published human facial expression recognition methods.This paper uses the pre-trained facial expression recognition model to design the prototype software of facial expression recognition,which can achieve good recognition accuracy by using the camera to capture facial expression in real time.
作者
顾状状
许学斌
路龙宾
豆阳光
GU Zhuangzhuang;XU Xuebin;LU Longbin;DOU Yangguang(School of Computer Science&Technology,Xi'an University of Posts&Telecommunications,Xi'an 710000;College of Geological Engineering and Geomatics,Chang'an University,Xi'an 710054)
出处
《计算机与数字工程》
2023年第10期2425-2430,共6页
Computer & Digital Engineering