计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (10): 199-204.DOI: 10.3778/j.issn.1002-8331.1802-0226

• 图形图像处理 • 上一篇    下一篇

多视角级联回归模型人脸特征点定位

贾项南,于凤芹,陈  莹   

  1. 江南大学 物联网工程学院,江苏 无锡 214122
  • 出版日期:2019-05-15 发布日期:2019-05-13

Multi-View and Cascaded Regression Model for Face Alignment

JIA Xiangnan, YU Fengqin, CHEN Ying   

  1. School of Internet of Things Engineering, Jiangnan University, Wuxi, Jiangsu 214122, China
  • Online:2019-05-15 Published:2019-05-13

摘要: 针对人脸姿态偏转较大导致人脸特征点定位精度低的问题,提出了多视角人脸特征点定位算法,采用随机森林局部学习与全局线性回归相结合的级联姿态回归(Cascaded Pose Regression,CPR)人脸特征点定位模型,在不同的人脸姿态视角下建立不同的模型,以多模型代替单一模型来提高人脸特征点定位的精度。首先采用CPR模型对不同视角下的人脸建立不同的模型;然后采用多视角生成模型(Multi-View Generative Model,MVGM)来评估输入人脸图片的姿态;最后根据评估的姿态选择相对应的模型,进而实现特征点的精确定位。仿真实验结果表明,相比于现有的几种人脸特征点定位算法,所提算法实现了更精确的定位效果。

关键词: 人脸特征点定位, 级联姿态回归, 随机森林, 全局线性回归, 多视角生成模型

Abstract: Aiming at the problem of low precision when locating facial landmarks due to large pose variations in face images, a multi-view face alignment algorithm is proposed. Cascaded Pose Regression(CPR) model is used to establish many different models under multi-view face images, which combines local learning principle with random forest and global linear regression. Multi-view models are established to improve face alignment accuracy, replacing a single model. Firstly, CPR model is used to establish different models for multi-view face images. Then, multi-view generative model is used to estimate the face pose of an input face image. Finally, according to face pose, a corresponding model is selected for an input image, which achieves high precision for face alignment. The experimental results show that the proposed face alignment algorithm has higher location precision than several existing face alignment algorithms.

Key words: face alignment, cascaded pose regression, random forest, global linear regression, multi-view generative model