Eigenfaces vs. Fisherfaces: Recognition Using Class Specific Linear Projection
IEEE Transactions on Pattern Analysis and Machine Intelligence
Shape Indexing Using Approximate Nearest-Neighbour Search in High-Dimensional Spaces
CVPR '97 Proceedings of the 1997 Conference on Computer Vision and Pattern Recognition (CVPR '97)
Real-Time and Markerless Vision-Based Tracking for Outdoor Augmented Reality Applications
ISAR '01 Proceedings of the IEEE and ACM International Symposium on Augmented Reality (ISAR'01)
Fully Automated and Stable Registration for Augmented Reality Applications
ISMAR '03 Proceedings of the 2nd IEEE/ACM International Symposium on Mixed and Augmented Reality
Pattern Classification (2nd Edition)
Pattern Classification (2nd Edition)
Visual Modeling with a Hand-Held Camera
International Journal of Computer Vision
Distinctive Image Features from Scale-Invariant Keypoints
International Journal of Computer Vision
Scene Modelling, Recognition and Tracking with Invariant Image Features
ISMAR '04 Proceedings of the 3rd IEEE/ACM International Symposium on Mixed and Augmented Reality
Randomized Trees for Real-Time Keypoint Recognition
CVPR '05 Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05) - Volume 2 - Volume 02
A Performance Evaluation of Local Descriptors
IEEE Transactions on Pattern Analysis and Machine Intelligence
PCA-SIFT: a more distinctive representation for local image descriptors
CVPR'04 Proceedings of the 2004 IEEE computer society conference on Computer vision and pattern recognition
An analysis-by-synthesis camera tracking approach based on free-form surfaces
Proceedings of the 29th DAGM conference on Pattern recognition
| Hi-index | 0.00 |
This paper shows an approach for automatic learning of efficient representations for robust image features. A video sequence of a 3D scene is processed using structure-from-motion algorithms, which provides a long validated track of robust 2D features for each tracked scene region. Thus each tracked scene region defines a class of similar feature vectors forming a volume in feature space. The variance within each class results from different viewing conditions, e.g. perspective, lighting conditions, against which the feature is not invariant. We show on synthetic and on real data that making use of this class information in subspace methods, a much sparser representation can be used. Furthermore, less computational effort is needed and more correct correspondences can be retrieved for efficient computation of the pose of an unknown camera image than in previous methods.