Robust Active Appearance Model Based Upon Multi-linear Analysis against Illumination Variation

Authors:
Gyeong-Sic Jo;Hyeon-Joon Moon;Yong-Guk Kim
Affiliations:
School of Computer Engineering, Sejong University, Seoul, Korea;School of Computer Engineering, Sejong University, Seoul, Korea;School of Computer Engineering, Sejong University, Seoul, Korea
Venue:
UAHCI '09 Proceedings of the 5th International on ConferenceUniversal Access in Human-Computer Interaction. Part II: Intelligent and Ubiquitous Interaction Environments
Year:
2009

Citing 7
Cited 0

A Multilinear Singular Value Decomposition

SIAM Journal on Matrix Analysis and Applications
From Few to Many: Illumination Cone Models for Face Recognition under Variable Lighting and Pose

IEEE Transactions on Pattern Analysis and Machine Intelligence
Active Appearance Models

IEEE Transactions on Pattern Analysis and Machine Intelligence
Multilinear Analysis of Image Ensembles: TensorFaces

ECCV '02 Proceedings of the 7th European Conference on Computer Vision-Part I
Interpreting Face Images Using Active Appearance Models

FG '98 Proceedings of the 3rd. International Conference on Face & Gesture Recognition
Active Appearance Models Revisited

International Journal of Computer Vision
Eigenfaces for recognition

Journal of Cognitive Neuroscience

Quantified Score

Hi-index	0.00

Visualization

Abstract

Independent Active Appearance Model (AAM) has been widely used in face recognition, facial expression recognition, and iris recognition because of its good performance. It can also be used in real-time system application since its fitting speed is very fast. When the difference between the input image and the base appearance of AAM is small, the fitting is smooth. However, when the difference can be large because of illumination and/or pose variation in the input image, the fitting result is unsatisfactory. In this paper, we propose a robust AAM using multi-linear analysis, which can make an Eigen-mode within the tensor algebra framework. The Eigen-mode can represent the principal axes of variation across the order of tensor and it can apply to AAM for increasing robustness. In order to construct both of original AAM and the present AAM, we employ YALE data base, which consists of 10 subjects, 9 poses, and 64 Illumination variations. The advantage of YALE data base is that we can use the coordinate of landmarks, which are marked for train-set, with ground truth. Because when the subject and the pose were same, the location of face isalso same. We present how we construct the AAM and results show that the proposed AAM outperforms the original AAM.