Matching Tensors for Pose Invariant Automatic 3D Face Recognition
Abstract
The face is an easily collectible and non-intrusive biometric used for the authentication and identification of individuals. 2D face recognition techniques are sensitive to changes in illumination, makeup and pose. We present a fully automatic 3D face recognition algorithm that overcomes these limitations. During the enrollment, 3D faces in the gallery are represented by third order tensors which are indexed by a 4D hash table. During online recognition, tensors are computed for a probe and are used to cast votes to the tensors in the gallery using the hash table. Gallery faces are ranked according to their votes and a similarity measure based on a linear correlation coefficient and registration error is calculated only for the high ranked faces. The face with the highest similarity is declared as the recognized face. Experiments were performed on a database of 277 subjects and a rank one recognition rate of 86.4% was achieved. Our results also show that our algorithm’s execution time is insensitive to the gallery size.
Cite
Text
Mian et al. "Matching Tensors for Pose Invariant Automatic 3D Face Recognition." IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2005. doi:10.1109/CVPR.2005.490Markdown
[Mian et al. "Matching Tensors for Pose Invariant Automatic 3D Face Recognition." IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2005.](https://mlanthology.org/cvpr/2005/mian2005cvpr-matching/) doi:10.1109/CVPR.2005.490BibTeX
@inproceedings{mian2005cvpr-matching,
title = {{Matching Tensors for Pose Invariant Automatic 3D Face Recognition}},
author = {Mian, Ajmal S. and Bennamoun, Mohammed and Owens, Robyn A.},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year = {2005},
pages = {120},
doi = {10.1109/CVPR.2005.490},
url = {https://mlanthology.org/cvpr/2005/mian2005cvpr-matching/}
}