一种基于机器学习的人脸情绪识别方法研究-《计算机技术与发展》

文章信息/Info

作者:: 李广鹏¹; 刘波¹; 李坤²; 黄思琦³; 1. 国防科技大学计算机学院,湖南长沙 410073;2. 国防大学政治学院教研保障中心信息技术室,上海 201602;3. 大连交通大学软件学院,辽宁大连 116000

Author(s):: LI Guang-peng¹; LIU Bo¹; LI Kun²; HUANG Si-qi³; 1. School of Computer,National University of Defense Technology,Changsha 410073,China;2. Information Technology Office of Teaching and Research Security Center,Political College of National Defense University,Shanghai 201602,China;3. School of Software,Dalian Jiaotong University,Dalian 116000,China

摘要:: 文中对人脸情绪识别整个过程所使用的算法进行分析和总结,并参考了国内外的人脸情绪识别方面的论文和报告,提出一种传统的基于机器学习的人脸图像情绪识别方法。文中将重点放在人脸表情核心部位(眼睛、眉毛、鼻子和嘴巴),将这些部位的特征向量作为人脸表情主要特征进行处理。使用的人脸关键部位特征提取算法是基于回归的 Explicit Shape Regression 算法,该算法在人脸特征部位定位方面取得了很好的效果。因为 Gabor 小波有很好的仿生效果,可以很好地表达表情的变化,所以文中的特征提取算法使用的是 Gabor 小波。但是考虑到提取出的特征维数过高,算法复杂,耗时较大,在后期用 PCA 算法进行降维处理,降维之后进行表情分类。

Abstract:: We analyze and summarize the algorithms used in the whole process of face emotion recognition,and propose a traditional face image emotion recognition method based on machine learning by referring to domestic and foreign papers and reports on face emotion recognition. We focus on the core parts of facial expression (eyes,eyebrows,nose and mouth) to deal with the feature vectors of these parts as the main features of facial expression. The Explicit Shape Regression algorithm based on regression is used as the key facial feature extraction algorithm,which has been proved to be very effective in face feature location. Because Gabor wavelet has great bionic effect,it can express the change of expression very well,so the feature extraction algorithm in this paper uses Gabor wavelet. But considering thatthe extracted feature dimension is too high,the algorithm is complex and time-consuming,the PCA algorithm is used to reduce the dimension in the later stage,and then the facial expression is classified after dimensionality reduction.