基于线性等距映射面部表情非线性流形估计  

Nonlinear Manifold Estimation of Facial Expression Based on Linear Isometric Mapping

作　　者：王艳[1] 张忠波[2] 

机构地区：[1]山西财经大学应用数学学院,太原030006 [2]吉林大学数学学院,长春130012

出　　处：《控制工程》2017年第1期124-129,共6页Control Engineering of China

摘　　要：为改善面部表情研究领域算法的识别效果,提出基于局部线性等距映射面部表情非线性流形估计算法。首先,针对多元冗余信息处理所导致的计算复杂度大幅增加的问题,利用等距映射过程(ISOMAP)和局部表情的线性嵌入过程(LLE)构建新的数据降维方法,可对高维的输入数据进行保守的近邻无监督训练;其次,利用Isomap和LLE实现输入数据的低维坐标映射,实现面部表情的局部聚类识别;最后,通过在Frey Raw数据库测试表明,所提方法在面部表情的非线性流形结构识别率及计算时间指标上要优于选取的对比算法,可有效对人脸表情及文本文字等信息进行处理。In order to improve the recognition effect of facial expression research, a nonlinear manifoldestimation algorithm based on local linear isometric mapping is proposed. Firstly, for a substantialincreasement in the multiple redundant information processing in computational complexity, here a new datadimension reduction method using isometric mapping （Isomap） and local expression of linear embedding（LLE） is constructed, which can be used for the conservative neighboring unsupervised training of highdimensional input data; Secondly, the Isomap and LLE are used to realize the low dimensional coordinatemapping of the input data, and the local clustering recognition of facial expressions is realized; Finally,according to FreyRaw test database, the provided method is superior to the selected contrast algorithm interms of the recognition rate and the computational time of the nonlinear manifold structure of facialexpression, which can effectively process the facial expression and text information.

关 键 词：非线性流形 面部表情 数据降维 估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]