基于表观模型的人脸特征点提取  

Facial feature point extraction based on active appearance model

作　　者：闫保中[1] 何伟 韩旭东 YAN Baozhong;HE Wei;HAN Xudong(College of Automation,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学自动化学院,黑龙江哈尔滨150001

出　　处：《应用科技》2020年第6期47-52,共6页Applied Science and Technology

摘　　要：针对在人脸特征点提取过程中模型预估能力不足,拟合过程中的非法形变导致的特征点提取速度慢和提取精度不高的问题,本文提出了一种基于主动表观模型(AAM)的人脸特征点提取算法的改进算法。将头部作初始分类,不同的头部姿态选取不同人脸模型进行拟合,这样能避免初始化模型与真实特征点位置相差过大,从而使得模型能更快的收敛,提高特征点提取速度。同时提出一种方法,对拟合过程中的形状变量加以限制,能有效过滤掉不满足人脸形状的特征点模型,防止拟合过程中的非法形变,使提取的人脸特征点更接近真实位置。实验结果表明,改进的基于AAM的人脸特征点提取算法在时间效率和准确率上都所有提高。In order to solve the problems of slow speed and low accuracy in the process of facial feature point extraction that are caused by insufficient model prediction ability and illegal deformation in the fitting process,an improved algorithm of facial feature point extraction is proposed in this paper based on active appearance model(AAM).The initial classification of head is carried out,and different head pose is fitted by different facial model.In this way,the difference of position between the initial model and the real feature points can be avoided so that the model can converge faster and the feature points can be extracted faster.Meanwhile,a method is proposed to filter out the features that do not meet the facial shape effectively through limiting the shape variables in the fitting process.The feature point model can prevent the illegal deformation in the fitting process,and make the extracted face feature point closer to the real position.The experimental results show that both time efficiency and accuracy of the improved facial feature point extraction algorithm are improved based on AAM.

关 键 词：表观模型 特征提取 人脸特征点 主动表观模型 非法变形 方向梯度直方图特征 拟合 头部姿态 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]