基于区域混合投影和支持向量机的人脸检测与定位  

Face Detection and Location Based on Region Hybrid Projection and Support Vector Machines

作　　者：徐从东[1] 罗家融[1] 舒双宝[1] 

机构地区：[1]中国科学院等离子体物理研究所

出　　处：《系统仿真学报》2008年第15期4031-4035,4040,共6页Journal of System Simulation

基　　金：中国科学院知识创新工程重要方向性项目(KJCX3.SYW.N4)

摘　　要：肤色信息是人脸的重要特征,根据肤色信息在YCbCr空间分布特点,提出区域混合投影概念,构造了图像的马氏距离图,在马氏距离图上用区域混合投影检测和定位人眼,确定可能的人脸区域。在此基础上以马氏距离图为输入构造支持向量机人脸检测器,对可能的人脸区域进行确认。实验表明,这种方法具有良好的人脸检测效果。Skin color information is an important characteristic of face image, According to skin color distributing in YCbCr color space, a novel concept region hybrid projection was proposed, and mahalanobis distance map of image was set up, Eyes were located by region hybrid projection and face area was fixed on mahalanobis distance map. Then, mahalanobis distance map served as input, face detector based on support vector machines was constructed, which validates face, Experimental result shows the approach to detect and locate face is effective.

关 键 词：肤色检测 马氏距离图 区域混合投影 支持向量机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]