DocumentCode
557791
Title
Multi-sample 3D face registration based on TPS transformation and linear combination model
Author
Hu, Yongli ; Zhu, Xianghui ; Sun, Yanfeng ; Yin, Baocai
Author_Institution
Beijing Key Lab. of Multimedia & Intell. Software Technol., Beijing Univ. of Technol., Beijing, China
Volume
3
fYear
2011
fDate
15-17 Oct. 2011
Firstpage
1343
Lastpage
1348
Abstract
To conquer the point-to-point registration of 3D face, we present an automatic registration method for multi-sample 3D face based on the thin-plate spline(TPS) transformation and the linear combination model. In the method, the data registration is divided into two stages. In the first stage, each sample face will be registered with the selected reference face by the TPS based non-rigid registration method to obtain the primary data registration; in the second stage, the combination model is set up based on the first stage primarily registered multi-sample data. For each sample face, matched with the combination model a similar reference face will be generated to facilitate the accurate data registration by TPS registration. To eliminate the outliers of different face samples, countermeasures are adopted in both stages of the data registration. The experiments on the BJUT-3D face database show that our method is effective for the registration of 3D face and robust to the outliers of face data.
Keywords
face recognition; image registration; splines (mathematics); BJUT-3D face database; automatic registration method; data registration; linear combination model; multisample 3D face registration; nonrigid registration method; point-to-point registration; reference face; thin-plate spline transformation; Accuracy; Data models; Databases; Face; Interpolation; Solid modeling; Three dimensional displays;
fLanguage
English
Publisher
ieee
Conference_Titel
Image and Signal Processing (CISP), 2011 4th International Congress on
Conference_Location
Shanghai
Print_ISBN
978-1-4244-9304-3
Type
conf
DOI
10.1109/CISP.2011.6100504
Filename
6100504
Link To Document