基于支持向量机分类器的活动形状模型  

Support vector machine classifier based active shape model

作　　者：杜春华[1] 杨杰[1] 张田昊[1] 吴证[1] 景旭[2] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200240 [2]西北农林科技大学信息工程学院,陕西杨凌712100

出　　处：《红外与激光工程》2006年第z4期469-476,共8页Infrared and Laser Engineering

基　　金：上海市科学技术委员会基金项目(03DZ14015)

摘　　要：为了精确地寻找活动形状模型中特征点的新位置,提出了一种基于支持向量机分类器的活动形状模型用于人脸特征点定位,即把寻找特征点新位置的任务转化为分类问题。起初,这是典型的两类分类问题,但两类分类器寻找特征点新位置效果并不理想。因此进一步提出把两类分类问题转化为多类分类问题。为每一个特征点训练一个支持向量机多类分类器,并用此分类器寻找该特征点新位置。实验结果表明,基于支持向量机多类分类器的活动形状模型比原始活动形状模型更为精确,稳健。In order to accurately find the new position for the landmark in Active Shape Model(ASM),Support Vector Machine(SVM)Based ASM is proposed to locate the facial landmarks,in which the problem of finding a new position for the landmarks is considered as an identification problem.Firstly,it is a typical two-class identification problem.However the accuracy of the finding via the two-class classifier is not satisfying.Then the above-mentioned two-class identification problem is further converted into a multi-class identification problem.Study a SVM classifier for each landmark and use it to find the new position for this landmark.Experimental results indicate the proposed SVM based ASM is more accurate and robust than the original ASM.

关 键 词：活动形状模型 支持向量机分类器 分类问题 新位置 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]