一种基于改进的ASM的人脸特征点定位方法  

Method of Human Facial Feature Points Positioning Based on Improved ASM

作　　者：韩玉峰[1] 王小林[2] 

机构地区：[1]安徽工业大学数理学院,马鞍山243032 [2]安徽工业大学计算机学院,马鞍山243032

出　　处：《计算机科学》2013年第4期271-274,共4页Computer Science

基　　金：国家自然科学基金项目(41075027);2011年安徽高校省级自然科学研究项目(KJ2011A040)资助

摘　　要：主动形状模型(ASM)在对目标点的搜索过程中,只采用了训练图像中标定点两侧法线方向的profile邻域内像素点的灰度信息,且对这些点等同视之;搜索时也只限于目标图像标定点两侧法线方向上的若干个像素,范围过于简单。考虑到彩色人脸图像的普及及其携带的丰富的信息,首先在RGB空间分三通道分别进行处理,其次对profile邻域内像素点赋予不同的权重,建立加权的局部灰度模型,最后把搜索空间拓展到包括传统法线和与之相互平行的相邻的两条法线上的像素点集上。仿真实验表明,以上3方面的改进大大提高了ASM方法对人脸特征点的定位精度,精度提高16.5%,是一种可行的改进方法。In the searching of ASM algorithm for target points,only surrounding local gray-level information of pixel points in the profile neighborhood on the training images is adopted,and the pixel points are equally treated.Furthermore,searching space is only limited to several pixel points along the normal direction on both sides of fixed point of target image.Obviously,the searching space is too simple.Considering that color facial images are easily obtained and have rich information,the paper suggested that firstly the images are processed on three channels of R,G and B,and secondly pixel points in the profile neighborhood are endowed with different weights on the construction of local gray-level model.Thirdly,searching space is extended to pixel points on traditional normal line and other two adjacent parallel normal lines.The results of experiments show that the new algorithm greatly improves the accuracy of positioning by sixteen point five percent than traditional algorithm,so the improving way is feasible.

关 键 词：主动形状模型(ASM) 局部灰度模型 人脸特征点定位 RGB空间 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]