The Beauty of Local Invariant Features The Beauty of Local Invariant Features Svetlana Lazebnik Svetlana Lazebnik Beckman Institute, University of Illinois at Urbana-Champaign Beckman Institute, University of Illinois at Urbana-Champaign IMA Recognition Workshop IMA Recognition Workshop University of Minnesota University of Minnesota May 22, 2006 May 22, 2006
21
Embed
The Beauty of Local Invariant Features Svetlana Lazebnik Beckman Institute, University of Illinois at Urbana-Champaign IMA Recognition Workshop University.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
The Beauty of Local Invariant FeaturesThe Beauty of Local Invariant Features
Svetlana LazebnikSvetlana LazebnikBeckman Institute, University of Illinois at Urbana-ChampaignBeckman Institute, University of Illinois at Urbana-Champaign
IMA Recognition WorkshopIMA Recognition Workshop
University of MinnesotaUniversity of Minnesota
May 22, 2006May 22, 2006
What are Local Invariant Features?What are Local Invariant Features?
• Descriptors of image patches that are invariant to certain classes of geometric and photometric transformations
Lowe (2004)
A Historical PerspectiveA Historical Perspective
ACRONYM: Brooks and Binford (1981)Alignment: Huttenlocher & Ullman (1987)Invariants: Rothwell et al. (1992)
Model-based methods: local shape, no appearance information
Appearance-based methods: global appearance, no local shape
– MSER and Hessian regions have the highest repeatability– Harris and Hessian regions provide the most correspondences– SIFT (GLOH, PCA-SIFT) descriptors have the highest performance
• 3D objects Moreels & Perona (2006)
– Features on 3D objects are much more unstable than on planar objects– All detectors and descriptors perform poorly for viewpoint changes > 30°– Hessian with SIFT or shape context perform best
– Laplacian regions with SIFT perform best– Combining multiple detectors and descriptors improves performance– Scale+rotation invariance is sufficient for most datasets
Comparative EvaluationsComparative Evaluations
Sparse vs. Dense Features: Sparse vs. Dense Features: UIUC texture datasetUIUC texture dataset
Lazebnik, Schmid & Ponce (2005)
25 classes, 40 samples each
Sparse vs. Dense Features: Sparse vs. Dense Features: UIUC texture datasetUIUC texture dataset
• A system with intrinsically invariant features can learn from fewer training examples
Invariant local features
Non-invariant dense patches
Baseline(global features)
SVM
Zhang, Marszalek, Lazebnik & Schmid (2005)
SVM
NN
NN
Multi-class classification accuracy vs. training set size
Sparse vs. Dense Features: Sparse vs. Dense Features: CUReT datasetCUReT dataset