摘要
在采用特征融合方法进行人脸表情识别时,通常会产生高维特征问题。针对这一问题,提出一种基于两步降维和并行特征融合的表情识别新方法。利用主成分分析法(principal component analysis,PCA)分别对待融合的两类特征在实数域进行第一次降维,将降维后的特征进行并行特征融合;为了解决在并行融合过程中产生的高维复特征问题,提出一种基于酉空间的混合判别分析方法(unitary-space hybrid discriminant analysis,unitary-space HDA)作为酉空间的特征降维方法。该方法是实数域混合判别分析法在酉空间内的扩展,并兼顾了复特征数据的类间判别信息及全局描述信息。对局部二值模式(local binary pattern,LBP)和Gabor小波特征进行融合,并在JAFFE和CK+表情数据集上开展对比实验。实验结果表明,该方法具有较好的高维复特征数据降维能力,并且有效提高了表情识别率。
When feature fusion method is used in emotion recognition, the problem of high dimensional features always ex- ists. In order to solve this problem, a novel facial expression recognition method is proposed in this paper, which is based on two-steps dimensionality reduction and parallel feature fusion. First of all, the PCA method is taken as the first-step fea- ture dimensionality reduction method for the two different feature vectors respectively in the real space. Afterward, the re- duced features are parallel fused in the unitary space. On the other hand, in order to solve the problem of high dimensional features which are generated in the process of parallel feature fusion, the unitary-space HDA method is proposed and is taken as the second-step feature dimensionality reduction method in the unitary space. It is the extension of the hybrid dis- criminant analysis method from the real space to the unitary space. Furthermore, this method combines both the complex between-class discriminant information and the complex global descriptive information of the parallel combined features. Several experiments are taken on the JAFFE and CK ＋ data sets, where local binary pattern features and Gabor wavelet fea- tures are fused. The experimental results indicate that the proposed method is capable of reducing high dimensional complex feature, and it also achieves higher recognition rate than the traditional feature fusion methods.
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2015年第3期377-385,共9页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
韩国科学与信息科技未来规划部2013年ICT研发项目(10039149)
重庆市自然科学基金(CSTC
2007BB2445)~~
