计算机技术与发展2017，Vol.27Issue(1)：30-33,38,5.DOI:10.3969/j.issn.1673-629X.2017.01.007
一种改进的三维局部约束模型初始化方法
An Improved Initialization Method for 3 D Constrained Local Model
摘要
Abstract
3D Constrained Local Model (CLM-Z) achieves high accuracy in detecting 3D facial landmarks (e. g. , eye corners,nose tip and mouth corners) via taking full advantage of both intensity and depth information. CLM-Z is conventionally initialized based on the location of face and mean 3D facial model. The effect of CLM-Z initialization on detection accuracy is evaluated quantitively by carrying out the following experiments:assessing the accuracy of CLM-Z with the conventional initialization method on the BU-4DFE database, translating the face to perturb the initial value of translation parameter,varying the size of detected face to perturb the initial value of scale parameter,varying the rotation angles around y-axis and z-axis to perturb the initial value of rotation parameter. Experimental results show that CLM-Z can tolerate translations up to approximately 1/6 of the width of the face,scalings between 0. 75 and 1. 50,and rota-tions within 20 degrees. Based on the above evaluation results,a novel initialization method is proposed further that exploits facial land-marks detected firstly on 2D texture images. Experiments on the BU-4DFE database show that the proposed initialization method can suc-cessfully improve the 3D landmark localization accuracy of CLM-Z approach.关键词
三维人脸特征点定位/三维局部约束模型/初始化/鲁棒级联姿态回归Key words
3D facial landmark localization/3D Constrained Local Model ( CLM-Z)/initialization/Robust Cascaded Pose Regression ( RCPR)分类
信息技术与安全科学引用本文复制引用
许进文,赵启军,陈虎..一种改进的三维局部约束模型初始化方法[J].计算机技术与发展,2017,27(1):30-33,38,5.基金项目
国家自然科学基金资助项目(61202160,61202161) （61202160,61202161）
科技部重大仪器专项(2013YQ49087904) （2013YQ49087904）
