基于模板匹配搜索的人脸特征点定位

Facial Feature Extraction Based on Template Matching Searching

针对人脸特征定位问题,利用Active Shape Model(ASM)定位的部分思想,提出一种实时的模板匹配搜索算法(Template Matching Searching,TMS).首先,对训练图像进行特征点标注,为每个特征点抽取一个矩形区域作为可能的匹配模板,并且对表示特征点位置的数据进行PCA处理,建立统计形状模型;然后,使用归一化相关为搜索图像的特征建立相应的检测器,在搜索图像过程中,检测器将为每个特征点产生一个响应,评估当前特征位置的优劣;最后,使用全局优化方法产生最优的特征点位置.结果表明,该算法可以鲁棒地对人脸特征点进行精确定位,并且克服了以往只能对完全正面人脸进行准确定位的缺陷.

This paper applied some ideas of the traditional ASM method aiming at the problem of facial feature extraction and presented a real-time algorithm called template matching searching （TMS）. At first, each feature is labeled on each training image with a rectangle area extracted to be the possible matching template. Then the data which accounts for the position of the features is processed using PCA and a statistical shape model is constructed. For the search image, the detectors for each facial feature are found using normalized correlation, with which the corresponding response for each facial feature is generated to evaluate the quality of the current position of the feature in the process of searching the image. A global optimization method is used to predict the most ideal position of the features. The experiments demonstrate that the algorithm for the facial feature extraction is precise and robust and overcomes the defect that the features can be accurately extracted only when searching the completely frontal face images.