节点文献

条件主动外观模型下的人脸特征点跟踪

Face Features Tracking with Conditional Active Appearance Model

  • 推荐 CAJ下载
  • PDF下载
  • 不支持迅雷等下载工具,请取消加速工具后下载。

【作者】 陈莹艾春璐

【Author】 Chen Ying1,2) and Ai Chunlu1) 1)(Key Laboratory o f Advanced Process Control for Light Industry(Ministry o f Education),Jiangnan University,Wuxi 214122) 2)(Key Laboratory o f System Control and In f ormation Processing(Ministry o f Education),Shanghai Jiao Tong University,Shanghai 200240)

【机构】 江南大学轻工过程先进控制教育部重点实验室上海交通大学系统控制与信息处理教育部重点实验室

【摘要】 为了完成人脸关键特征点的精确定位跟踪,提出一种改进的基于反向合成匹配算法的条件主动外观模型匹配及其初始化算法.该算法假设已知正面人脸的关键特征点,首先通过建立散乱点对应与标定点对应之间的映射,根据给定的正面人脸标定点对任意姿态的侧面人脸进行自动初始标定,映射关系由核岭回归算法学习得到;将该标定点作为人脸跟踪算法的初始化点,然后利用条件主动外观模型反向合成匹配算法建立正面与任意姿态人脸的外观和形状模型,并对模型参数进行迭代优化;最后得到最优的任意姿态人脸的轮廓点,完成人脸跟踪.实验结果证明,与同类方法相比,该算法表现出了良好的性能,可在较短的计算时间内获得较高的定位精度.

【Abstract】 An improved active appearance model(AAM) based on inverse compositional algorithm,called conditional active appearance model(CAAM),is proposed to complete the facial key feature points localization and tracking accurately.First,the mapping between the scattered correspondence and the structured correspondence is established via kernel ridge regression.Assuming the annotation of frontal face image is known,the feature points of profile face images are automatically initialized from the frontal image based on the mapping.Second,the shape model and the appearance model between the frontal face and arbitrary profile face are established with conditional active appearance model,and the model parameters are optimized iteratively through inverse compositional algorithm.Finally,face features tracking is completed by achieving shape contour with optimum conditional model parameters.It is proved by experiments that the method demonstrates good performance in face tracking.Higher accuracy in positioning and less computing time can be obtained compared with other related methods.

【基金】 国家自然科学基金(61104213);江苏省自然科学基金(BK2011146);上海交通大学系统控制与信息处理教育部重点实验室开放基金(SCIP2011008)
  • 【文献出处】 计算机辅助设计与图形学学报 ,Journal of Computer-Aided Design & Computer Graphics , 编辑部邮箱 ,2013年04期
  • 【分类号】TP391.41
  • 【被引频次】27
  • 【下载频次】285
节点文献中: 

本文链接的文献网络图示:

本文的引文网络