基于特征人脸和肤色统计的人脸检测  

Human Face Detection Using Eigen-face and Skin Color

作　　者：潘志庚[1,2] 邹鹏程[1] 梁荣华[1] 

机构地区：[1]杭州电子工业学院虚拟现实与多媒体实验室 [2]浙江大学CAD&CG国家重点实验室,杭州310027

出　　处：《系统仿真学报》2004年第6期1346-1349,共4页Journal of System Simulation

基　　金：浙江省自然科学基金(602134) ;浙江省自然科学基金优秀人才基金项目(RC40008)共同资助。

摘　　要：提出一种在视频中检测人脸的新方法,该方法首先用特征人脸检测第一帧画面中人脸的位置和大小,将它作为卡尔曼(Kalman)滤波的初始值和肤色分布统计的训练样本,然后利用卡尔曼滤波预测人脸,最后用肤色分布模型确定人脸,就可以检测到人脸。这种方法的优点是可以在复杂背景下快速准确地检测到人脸位置和大小。实验结果表明使用该方法可以实时准确地检测到视频中的人脸。In this paper, a new method for human face detection from video sequence is presented. First, the algorithm makes full use of the eigenface to detect the position and the size of the first frame face which are used for initial value of Kalman filter and the training sample of skin color. Secondly, Kalman filter is used to predict the position and the size of human face. Finally, the position and the size of human face are obtained by the skin color model and then human face is detected. The advantage of the algorithm is that the position and the size of human face are detected in the complex background quickly and exactly. Experimental results show that human face is exactly detected in realtime in the video sequence with our approach.

关 键 词：肤色模型 特征人脸 肤色过滤 卡尔曼滤波 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]