计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (20): 154-160.DOI: 10.3778/j.issn.1002-8331.1604-0141

• 模式识别与人工智能 • 上一篇    下一篇

一种用于脸部特征检测的分层概率模型研究

石贵民1,2,余文森1,2   

  1. 1.武夷学院 数学与计算机学院,福建 武夷山 354300
    2.认知计算与智能信息处理福建省高校重点实验室,福建 武夷山 354300
  • 出版日期:2017-10-15 发布日期:2017-10-31

Research on hierarchical probability model for facial feature detection

SHI Guimin1,2, YU Wensen1,2   

  1. 1.School of Mathematics and Computer, Wuyi University, Wuyishan, Fujian 354300, China
    2.The Key Laboratory of Cognitive Computing and Intelligent Information Processing of  Fujian Education Institutions, Wuyishan, Fujian 354300, China
  • Online:2017-10-15 Published:2017-10-31

摘要: 脸部特征检测问题是计算机视觉领域的研究热点。因为脸部外观和形态随着条件的变化而变化,因此面部特征检测比较复杂。针对现有脸部特征检测算法的不足,提出一种已知图像测量数据后能够推断出真实脸部特征位置的分层概率模型。针对每个脸部子部位的局部形态变化进行间接建模;通过搜索模型的最优结构和参数设置,在更高层次上学习脸部子部位、脸部表情和姿态间的联合关系。该模型综合利用了脸部子部位自下而上的形态约束以及脸部子部位间自上而下的关系约束来推断出脸部特征的真实位置。利用基准数据库进行了仿真实验。实验结果表明,该方法的检测性能要明显优于目前最新的人脸特征检测算法。

关键词: 脸部特征检测, 分层概率模型, 局部形态, 约束, 检测误差

Abstract: Facial feature detection problem is a hot research topic in the field of computer vision. It is a nontrivial task since the appearance and shape of the face tend to change under different conditions. Aiming at the deficiency of the existing facial feature detection algorithms, a hierarchical probabilistic model is proposed that can infer the true facial feature locations given image measurements. Firstly, it implicitly models the local shape variation for each facial component. Secondly, it learns the joint relationship among facial components, the facial expression and the pose in the higher level by searching the optimal structure and parameterizations of the model. Finally, the true facial feature locations are inferred through the bottom-up lower level shape constraints of facial components and the top-down constraint from the relationship among facial components. Experimental results on benchmark databases demonstrate that, the detection performance of the proposed method is obviously better than that of the most recent face feature detection algorithm.

Key words: facial feature detection, hierarchical probabilistic model, local shape, constraints, detection error