基于LLE的彩色图象人脸检测  

Face Detection of Color Image Based on LLE

作　　者：吴俊强[1] 周激流[1] 何坤[1] 

机构地区：[1]四川大学电子信息学院,成都610064

出　　处：《四川大学学报（自然科学版）》2006年第4期796-800,共5页Journal of Sichuan University(Natural Science Edition)

摘　　要：分析了人脸与非人脸之间的本质区别,提出了运用局部线形嵌入(LLE)的非线性降维方法,解决非线性结构的高维数据(图象)低维表示的问题,实现了高维输入数据点映射到一个全局低维坐标系,同时保留了邻接点之间的空间关系(即高维空间的几何结构).此算法不仅能够有效地发现数据的非线性结构,同时还具有平移、旋转不变性.运用LLE算法对图象进行降维,再对降维后的数据运用支持向量机(SVM)分类器进行人脸和非人脸的分类.实验结果表明,该人脸检测方法测率较高,并且不受姿态、表情和光照的影响.The locally linear embedding （LLE） algorithm is presented to reduce the dimensionality of image data. The locally linear embedding algorithm can reveal the intrinsic distribution of data, which cannot be provided by classical linear dimensionality reduction methods. The nonlinear structure in high dimensional data space was exploited with the local symmetries of linear reconstructions. The data points in high dimensional space were mapped into corresponding data points in lower dimensional space under preserving distance between data points. And it also has invariability in translation and rotation. Moreover, a new face detection method using support vector machine （SVM） based on LLE is proposed. The SVM classifier is trained by the data whose dimensionality is reduced by LLE and then can differentiate between face and non-face which is encoded by LLE. Experiment result shows that, detection rate is high and the method is not affected by poses, expression and illumination.

关 键 词：局部线性嵌入 LLE 非线性降维 支持向量机 人脸检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]