多姿态人脸照片的姿态估计  

Pose Estimating of Pose-Varied Face Image

作　　者：张小平[1] 朱红锋[1] 刘志镜[1] 

机构地区：[1]西安电子科技大学计算机信息应用研究中心,陕西西安710071

出　　处：《计算机仿真》2005年第4期202-205,共4页Computer Simulation

摘　　要：提出了一种基于面部图像的新的匹配系统。在这个系统中,输入的图像与各种人脸姿态的数据库图像进行比较,然后,匹配的图像给出了人脸姿态。图像数据库不仅包括各种人脸姿态,而且也包括不同的光照条件,如此,这个人脸姿态评价系统适用于不同的光照条件。对于收集各种不同面部图像,这里是通过计算机自动产生,而不是拍摄实际的照片。特征空间方法被用于寻找与输入面部图像匹配的图像。因为不同的光照图像被收集在面部图像数据库中,故提取的主特征向量主要依靠人脸姿态。由于通过选用主特征向量而减少了向量的维数,故这个匹配过程是很快的。这个姿态评价系统能够继续跟踪在不同的光照条件下不同人的人脸姿态。A new system for estimating face from a facial image is proposed. In t his system, input facial image is compared with database of images of various fa ce pose, then the matched image provides the face pose. The database of images i ncludes not only various face poses but also various illumination conditions, so that the face pose estimating system can be used under various illumination cond itions. For collecting such various facial images, they are generated by compute r , rather than taking real images. Eigen space method is used for searching the m atched image with input facial image. Since various illumination images are coll e cted in the database of facial images, the extracted principle eigen vectors mos tly depend on the face pose. The matching process is also fast because it is pe rformed in small dimensional space spanned by only selected eigen vectors. The p roposed pose estimating system can continuously track the face pose of different person under different light conditions.

关 键 词：人脸识别 多姿态 姿态估计 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]