基于多模板匹配的自动人脸检测

本文提出了正面人脸检测的一种新算法.不同于以往将人脸看作高维空间中的单一聚类的方法,本文提出,人脸在高维空间中以多聚类形式存在.并采用最优聚类的方法将人脸分为四个聚类.以每个聚类的中心为一个模板,通过基于马氏距离(Mahalanobis distance)的多模板匹配法,检测出图像中的人脸.实验结果表明,该方法优于基于单一聚类的方法.

Multiple Template Matching for Frontal-View Face Detection

In this paper,we present a new method for frontal view face detection.Instead of viewing face patterns as a single cluster in the high dimensional space as several previous approaches,we propose that face patterns can be better described by multiple clusters.We obtain four clsters by employing an optimal clustering method.Using each cluster center as a template,we adopt Mahalanobis distance based template matching to find the face location in an image.The experimental results show that our method outperforms the methods that are based on a single cluster.