Numerical recipes in C (2nd ed.): the art of scientific computing
Numerical recipes in C (2nd ed.): the art of scientific computing
What Is the Set of Images of an Object Under All Possible Illumination Conditions?
International Journal of Computer Vision
From Few to Many: Illumination Cone Models for Face Recognition under Variable Lighting and Pose
IEEE Transactions on Pattern Analysis and Machine Intelligence
A signal-processing framework for inverse rendering
Proceedings of the 28th annual conference on Computer graphics and interactive techniques
Lambertian Reflectance and Linear Subspaces
IEEE Transactions on Pattern Analysis and Machine Intelligence
Ridge Regression Learning Algorithm in Dual Variables
ICML '98 Proceedings of the Fifteenth International Conference on Machine Learning
The CMU Pose, Illumination, and Expression (PIE) Database
FGR '02 Proceedings of the Fifth IEEE International Conference on Automatic Face and Gesture Recognition
Acquiring Linear Subspaces for Face Recognition under Variable Lighting
IEEE Transactions on Pattern Analysis and Machine Intelligence
IEEE Transactions on Pattern Analysis and Machine Intelligence
Orthogonal Laplacianfaces for Face Recognition
IEEE Transactions on Image Processing
Convergence Analysis of Affinity Propagation
KSEM '09 Proceedings of the 3rd International Conference on Knowledge Science, Engineering and Management
An incremental affinity propagation algorithm and its applications for text clustering
IJCNN'09 Proceedings of the 2009 international joint conference on Neural Networks
Hi-index | 0.00 |
Previous work has shown that human faces under variable lighting conditions can be modeled by low-dimensional subspaces called illumination subspaces that can be computed using images under a universal lighting configuration. This configuration can be estimated using Harmonic images. However, harmonic images can only be obtained by using 3D information, and thus can be restrictive. In this paper, we overcome this limitation by presenting a completely data-driven method to find good universal lighting configurations. Motivated by the fact that affinity propagation clustering finds the cluster centers from the real images, we use affinity propagation clustering on real images taken under variable lighting conditions to find the cluster centres and use them to determine the lighting configuration. The illumination subspace for each individual is spanned by their images acquired in this lighting configuration. Matching is performed by comparing the distances to these individual illumination subspaces. Further, kernel methods are used to explore the non-linear structures of the illumination cone and carry out the illumination subspace methods in the kernel induced feature space. Experiments conducted on the Extended Yale Face B database demonstrate that the configuration obtained by our method is better than earlier recommended configurations. We also demonstrate that our technique is robust to pose variations using the CMU PIE database.