摘要
针对局部二值模式(LBP)对非一致性光照敏感和随机噪声不稳健等问题,提出融合完整局部二值模式(CLBP)与几何显著特征的情感识别算法。利用Dlib库提取面部68个特征点,通过选取显著特征点计算中立及巅峰时刻欧几里德距离,构建特征比例向量。采用CLBP提取面部细粒度纹理特征,串联融合其特征直方图,利用随机森林进行表情分类。实验表明该算法在CK+数据集上具有92.8%的识别准确率,优于传统表情识别算法。
Aiming at the problems of local binary pattern(LBP)being sensitive to non-uniform illumination and random noise instability,this paper proposes an emotion recognition algorithm that combines complete local binary pattern(CLBP)with geometrically significant features.Firstly,68 feature points of the face were extracted using the Dlib library,and the feature vector was calculated by selecting salient feature points to calculate the Euclidean distance at the neutral and peak times.Then use CLBP to extract facial fine-grained texture features,and then fuse the feature histograms in series to use random forest for facial expression classification.Experiments show that the algorithm has a recognition accuracy of 92.8%on the CK+dataset,which is better than the traditional expression recognition algorithm.
作者
王从澳
黄润才
孙延标
杨彬
孙刘成
WANG Congao;HUANG Runcai;SUN Yanbiao;YANG Bin;SUN Liucheng(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201600,China)
出处
《智能计算机与应用》
2020年第5期52-55,共4页
Intelligent Computer and Applications
关键词
CLBP
纹理特征
随机森林
表情分类
CLBP
texture features
random forest
expression classification
