摘要
为提高多视角人脸表情的识别效果,提出一种基于局部保留投影的GAN多视角人脸表情识别方法。通过在判别空间上添加不相关的约束提高分类性能,利用随机森林算法完成人脸表情的粗分类;对于每个子类别,利用生成对抗网络(GAN)的博弈思想设计特征提取器、特征合成器和判别器,通过判别器与特征提取器之间的对抗训练提高分类准确率。实验结果表明,该方法在脸部左侧的识别率高于脸部右侧,精细分类效果明显优于粗分类。相比其它方法,所提方法的识别效果最佳,当视角为0°时,在BU3DFE数据集和CK+数据集上的识别准确率分别是98.97%和97.58%。
To improve the recognition effects of multi view facial expression,a multi view facial expression recognition method based on local preserving projection was proposed.Irrelevant constraints were added to the discriminant space to improve the classification performance,and random forest algorithm was used to complete the rough classification of facial expressions.For each subcategory,a feature extractor,a feature synthesizer and a discriminator were designed based on the game theory of gene-rated antagonism network(GAN).Through the confrontation training between discriminator and feature extractor,the classi-fication accuracy was improved.The results show that the recognition rate of the proposed method on the left side of the face is higher than that on the right side of the face,and its fine classification effect is significantly better than that of the coarse classi-fication.Compared with other contrast methods,the recognition accuracy of the proposed method is 98.97%and 97.58%on BU3DFE data set and CK+data set respectively when the angle of view is 0°.
作者
夏辉丽
王秉政
XIA Hui-li;WANG Bing-zheng(College of Computer and Artificial Intelligence,Zhengzhou University of Economics and Business,Zhengzhou 451191,China;College of Computer and Communication Engineering,Zhengzhou University of Light Industry,Zhengzhou 450002,China)
出处
《计算机工程与设计》
北大核心
2023年第6期1781-1788,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(61872126)
河南省重点研发与推广专项(科技攻关)基金项目(232102210085)。
关键词
局部保留投影
生成对抗网络
粗分类
多视角
人脸表情识别
随机森林
判别空间
local preserving projection
generated antagonism network
rough classification
multi view
facial expression recognition
random forest
discriminant space
