视频人脸的器官快速定位  

Fast apparatus location in video face

作　　者：段雨梅[1] 周激流[1] 何坤[1] 刘昶[1] 

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《四川大学学报（自然科学版）》2009年第1期61-64,共4页Journal of Sichuan University(Natural Science Edition)

基　　金：四川省青年软件创新工程(201)

摘　　要：传统的人脸器官定位算法系统运算时间较长,不适于视频人脸的器官定位。运用了Adaboost从视频图像中检测出人脸区域并确定其位置,将人脸区域进行肤色分割并确定出器官位置:首先对视频相邻两帧的人脸肤色区域差分初步确定眼部区域,再根据人脸器官的几何分布特征精确的定位眼部区域。最后利用伪H分量的数据分布确定嘴唇区域。通过实验表明本文的视频人脸检测算法处理时间较短和定位准确率较高。Traditional face apparatus location need more time to compute, it is not fit for apparatus location in video face. The authors use Adaboost algorithm to detect the face region from video images and get the position, divide the skin region from the face region and then do apparatus location as： first ,the authors can find the eye region by getting the difference of image from the two adjacent face skin regions, then the authors can confirm the exact eye region depending on the distribution character of the apparatus . At last, the authors use the distribution in pseudo hue level to divide the lip region from the face. Proven by the experiments, the method in this paper could detect the apparatus region quickly and precisely.

关 键 词：视频人脸 器官定位 ADABOOST 伪色调分量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]