基于DE-Gabor特征的人脸表情识别

Facial Expression Recognition Based on DE-Gabor Feature

种族、性别等个体身份的差异在面部表情识别过程中恒定存在的,会降低系统的分类性能。为此,本文提出一种DE-Gabor特征增强方法的身份鲁棒性。首先针对Gabor特征的高维问题,提出双重下采样策略进行降维,获得紧凑的E-Gabor特征。然后针对身份信息的干扰,将E-Gabor表情特征分别在中性特征字典和表情特征字典上进行协同稀疏表示,构建样本的虚拟中性特征和虚拟表情特征,两者差分编码获得增强身份独立性的DE-Gabor特征。最后,基于DEGabor特征训练SVM模型进行表情分类。此外,将DE-Gabor用于不同种族、不同性别的数据集,探究不同文化背景下身份干扰下表情识别之间的规律。在BU3DFE数据集上的实验结果表明:DE-Gabor特征的分类性能优于其它方法。

Individual identity differences such as race and gender always exist in the process of facial expression recognition,which will reduce the classification performance of the system.Therefore,this chapter proposes anDE-Gabor feature to enhance the robustness of identity.Firstly,a double subsampling strategy is proposed to reduce the dimension of Gabor feature to obtain a more compact E-Gabor feature.Then,according to the interference of identity information,the E-Gabor features were sparsely represented on neutral feature dictionary and expression feature dictionary respectively,and the virtual neutral feature and virtual expression feature of the sample were reconstructed,and the two differential coding were used to obtain independent identity features.Finally,the SVM model was trained based on DE-Gabor features to classify facial expressions.In addition,the DE-Gabor featurewere applied to data sets of different races and genders to explore the relationship between identity interference and facial expression recognition under different cultural backgrounds.Experimental results on BU3DFE data set show that the classification performance of DE-Gabor features is better than other methods.