基于多姿态知识模型和模板的快速人脸检测  

Rapid Face Detection with Multi-pose Knowledge Models and Templates

作　　者：邵平[1] 杨路明[1] 黄海滨[1] 曾耀荣[2] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410000 [2]玉林师范学院物理与信息科学系,广西玉林537000

出　　处：《小型微型计算机系统》2007年第2期346-350,共5页Journal of Chinese Computer Systems

基　　金：广西教育厅面上项目(桂教科研[2004]20号)资助;玉林师范学院重点科研项目(05DWX16)资助;玉林师范学院科研项目(05YBWX23)资助.

摘　　要：提出了一种多姿态知识模型,并以之从人脸器官梯度图中获得候选脸的大小、位置、姿态类别和眼、嘴重心坐标,然后按姿态类别将候选脸与对应的模板进行匹配确认人脸.该人脸检测算法集人脸检测、姿态估计和眼、嘴定位于一体,具有检测速度快的特点,适于多姿态多人脸场合的人脸检测.该算法只利用了图像的灰度信息,因此对灰度图像和彩色图像的人脸检测均适用.Multi-pose knowledge models are proposed to acquire the candidate face information ot size, position, sort ot pose and the center coordinates of eyes and mouth from two-value grads image of face organs. According to the sort of pose, the candidate face is affirmed as a true face through template matching with corresponding templates. The face detection, pose estimation and orientation of eyes and mouth are integrated in the algorithm and the detection speed is quite fast. This algorithm is suitable for face detection with multi-pose and multi-face, and since only the gray information of image is used, it can be used for face detection in either gray image or color image.

关 键 词：人脸检测 梯度特征 姿态估计 模板匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]