基于小波和流形学习的人脸姿态表情分析  

FACIAL POSE AND EXPRESSION ANALYSIS BASED ON WAVELETS AND MANIFOLD LEARNING

作　　者：徐秀秀[1] 梁久祯[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机应用与软件》2015年第3期167-171,176,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61170121)

摘　　要：针对人脸研究领域中高维数据产生的计算复杂度问题,提出基于小波分解的流形学习方法,对高维数据进行降维,从而达到降低计算复杂度的目的。该方法对人脸图像进行不同层次的小波分解保留低频分量后再分别应用局部线性嵌入(LLE)及局部保持投影(LPP)两种流形学习算法。实验在Frey和CMU PIE人脸库上进行,给出人脸姿态和表情分布变化的实验结果,并分析了运行时间和经小波分解得到的低频子图像的能量。结果表明,基于小波分解的流形学习算法对于降低计算复杂度和保持图像信息是有效的。In order to solve the problem of computational complexity caused by the high-dimensional data in human faces research area,we propose a wavelet decomposition-based manifold learning method to reduce the dimensionality of high-dimensional data so as to achieve the goal of decreasing computational complexity. The method carries out wavelet decomposition in different levels on face images,and retains low frequency components,then applies two manifold learning algorithms,the Locally-Linear Embedding（ LLE） and the Locality Preserving Projections（ LPP）,respectively. Experiments are conducted on the Frey database and the CMU PIE database,the experimental result of distribution and changes of face pose and facial expression are presented,the runtime and the energy of low frequency sub-image derived from wavelet decomposition are analysed as well. Experimental results show that the proposed method is efficient in reducing computational complexity and image information preserving.

关 键 词：小波分解 局部线性嵌入 局部保持投影 数据降维 人脸姿态表情 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]