基于肤色马氏距离图的人脸检测  

Human Face Detection Based on Mahalanobis Distance Map to Skin Color

作　　者：舒双宝[1] 罗家融[1] 徐从东[1] 刘连忠[1] 

机构地区：[1]中国科学院等离子体物理研究所,230031

出　　处：《微计算机信息》2008年第16期268-269,306,共3页Control & Automation

基　　金：国家自然科学基金(10475079)

摘　　要：介绍利用肤色信息、特征脸与径向基概率神经网络相结合的方法进行人脸检测,能准确快速地检测出彩色图像中的人脸区域。首先构造皮肤YUV颜色模型,检测彩色图像中的皮肤区域,并计算出该区域相对于皮肤的马氏距离图;根据其马氏距离图利用主成分分析法构建特征脸子空间;构造出径向基概率神经网络,以马氏距离图在各特征脸子空间的投影系数为网络的输入,计算出该区域是否是人脸图像。实验证明,这种方法是有效的。A method of human face detection using eigenface is presented, which is based on the color of skin and radial basis probabilistic neural network （RBPNN）. It can detect the face parts from a color image exactly and fast. A YUV skin color model is constructed, which detects skin parts in a color image and figures out mahalanobis distance map to skin color. Based on mahalanobis distance map and using PCA , Eigenface subspace is found out and An RBPNN model is constructed. On the RBPNN, the input is coefficient vector obtained from mahalanobis distance map projecting on subspace, and the output determines whether this part is face image. Experimental results show the method is effective.

关 键 词：人脸检测 马氏距离图 主成分分析 径向基概率神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]