DocumentCode
615181
Title
A stable and accurate multi-reference representation for surfaces of R3: Application to 3D faces description
Author
Gadacha, W. ; Ghorbel, Faouzi
Author_Institution
GRIFT Res. Group, Nat. Sch. of Comput. Sci. (NSCS), La Manouba, Tunisia
fYear
2013
fDate
22-26 April 2013
Firstpage
1
Lastpage
5
Abstract
In this paper, a 3D surface representation defined around several reference points taken on the surface is introduced. Such representation is obtained from the superposition of a set of indexed levels of geodesic curves and radial lines. A sampling criterion through a generalized version of the Shannon theorem allows the determination of the minimum resolution of both curves that describes faithfully the surface. The choice of several reference points is done to overcome the instability due to landmarks detection errors. Thus, the robustness of the obtained representation under small reference points positions distortion is showed. Then, its accuracy according to pose variation in facial description field is studied. Experimentations are performed on the 3D facial database Bosphorus[8]. Results show also the discrimination power of such representation.
Keywords
face recognition; image representation; image sampling; information theory; pose estimation; visual databases; 3D faces description; 3D surface representation; Shannon theorem; discrimination power; facial database; generalized version; geodesic curves; indexed levels; landmarks detection errors; multireference representation; pose variation; radial lines; reference points positions distortion; sampling criterion; superposition; Accuracy; Databases; Face recognition; Nose; Robustness; Shape; Surface treatment; 3D surface representation; Hausdorff distance; Shannon theorem; facial surfaces; parametrisation; pose variation; reference points; robustness;
fLanguage
English
Publisher
ieee
Conference_Titel
Automatic Face and Gesture Recognition (FG), 2013 10th IEEE International Conference and Workshops on
Conference_Location
Shanghai
Print_ISBN
978-1-4673-5545-2
Electronic_ISBN
978-1-4673-5544-5
Type
conf
DOI
10.1109/FG.2013.6553820
Filename
6553820
Link To Document