Pose Angle Determination by Face, Eyes and Nose Localization
Pose Angle Determination by Face, Eyes and Nose Localization
复制标题
DOI：
10.1109/cvpr.2005.582
复制
发表时间：
2005-06
期刊：
影响因子：
--
通讯作者：
P. Sankaran;Satyanadh Gundimada;R. Tompkins;V. Asari
中科院分区：
文献类型：
--
作者：
P. Sankaran;Satyanadh Gundimada;R. Tompkins;V. Asari
FRGC的目标是开发利用人脸数据库中高质量图像的算法。本文中的算法就是基于这一点而开发的。我们提出了使用多视图分类器估计姿态的方法。基于姿态和面部几何知识，找到可能的眼睛位置感兴趣的区域。采用自适应阈值法提取出该区域内最暗的像素，形成眼睛区域。利用这些信息形成感兴趣的鼻子定位区域。使用梯度算子获得该区域的边缘。使用直方图阈值法在该区域找到鼻子。绘制线条连接每个区域的中心点。垂线绘制于连接眼睛中心的线。根据这些线的斜率确定姿态角。使用FRGC版本2实验号4图像集测试眼睛定位算法，在版本1图像集上测试鼻子定位和位姿角方法。
FRGC aims to develop algorithms that make use of the high quality images in the face database. The algorithms in this paper have been developed with this in view. We present methods to estimate pose using multi-view classifiers. Based on the knowledge of pose and face geometry a region of interest of possible eye locations is found. An adaptive thresholding is used to pick out the darkest pixels in this region, which forms the eye region. Using this information region of interest of nose location is formed. A gradient operator is used to obtain the edges in this region. Histogram thresholding is used to find the nose in this region. Lines are drawn joining the center points of each of the regions. Perpendicular is drawn to the line joining the eye centers. Based on the slopes of these lines the pose angle is determined. The FRGC version 2 experiment number 4 image set is used to test the eye location algorithm while the nose location and pose angle methods are tested on version 1 image set.