![Page 1: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/1.jpg)
Bachelor thesis previewDaniel Mewes
Extracting Point Features for Symmetry Detection
![Page 2: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/2.jpg)
Overview
● Problem description– Symmetry detection
– Feature extraction
● My approach– The framework
– Scene structure analysis
● Preliminary results
![Page 3: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/3.jpg)
Overview
● Problem description– Symmetry detection
– Feature extraction
● My approach– The framework
– Scene structure analysis
● Preliminary results
![Page 4: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/4.jpg)
Symmetry Detection
● Given a 3D scene● Detect repeating structures
Figure fromBokeloh et al. 2009
![Page 5: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/5.jpg)
Symmetry Detection
● Many useful applications, e.g.– Simultaneous editing
– Data compression
– Data reconstruction and refinement
![Page 6: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/6.jpg)
Symmetry Detection
● Input● Find subsets with corresponding
transformations● Rigid case: are rigid transformations
⊂ℝ3
S⊂f 1 , ... , f n :S
f 1 , ... , f n
![Page 7: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/7.jpg)
Features
● Symmetry Detection is computationally complex
● Techniques to reduce computational costs:– Downsampling
– Randomized approaches (RANSAC)
– Features
– Combinations of these
![Page 8: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/8.jpg)
Features
● Idea: – Preselect relevant subset of the data
– Find symmetry candidates in the subset
– Validate
● Simplest class: Point features
![Page 9: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/9.jpg)
Feature Extraction
● Goal for feature extraction for symmetry detection
– Problem:
➔ Consistent features across all instances
One instance misses a feature
No symmetry candidate
Undetectable symmetry
![Page 10: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/10.jpg)
Feature Extraction
● How to extract point features?
![Page 11: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/11.jpg)
Feature Extraction
● Related work: Slippage features(Bokeloh et al. 2008, Gelfand et al. 2004)
– Perform local slippage analysis
– Select local maxima
No slippable translations
Slippable to the “right”
Slippable in two translational dimensions
![Page 12: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/12.jpg)
Feature Extraction
● Example: Slippage features (Bokeloh et al. 2008, Gelfand et al. 2004)
non-slippable
one slippable motion
![Page 13: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/13.jpg)
Feature Extraction
● So far: Local criteria
Feature? Feature?
![Page 14: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/14.jpg)
Feature Extraction
● Turns out local criteria are insufficient sometimes
Feature? Feature?
![Page 15: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/15.jpg)
Overview
● Problem description– Symmetry detection
– Feature extraction
● My approach– The framework
– Scene structure analysis
● Preliminary results
![Page 16: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/16.jpg)
My Approach
● Interleave iteratively:– Feature extraction
– Partial symmetry detection
● Missing features ← reconstruct from other instances
![Page 17: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/17.jpg)
The Framework
● Input
![Page 18: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/18.jpg)
The Framework
● Interest Function: Assigns local scores to points● e.g. slippage analysis
![Page 19: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/19.jpg)
The Framework
● Selection Function: Select feature points based on per-point scores
![Page 20: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/20.jpg)
The Framework
● Propagation Function: analyze structure, guess about additional features
![Page 21: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/21.jpg)
The Framework
● Combiner Function: Incorporate guesses into per-point scores
![Page 22: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/22.jpg)
The Framework
● Select new set of feature points based on the combined scores
![Page 23: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/23.jpg)
The Framework
● Iterate: repeat propagation, combination and selection with new feature points...
![Page 24: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/24.jpg)
Overview
● Problem description– Symmetry detection
– Feature extraction
● My approach– The framework
– Scene structure analysis
● Preliminary results
![Page 25: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/25.jpg)
Scene Structure Analysis
● How to analyze the scene's structure?
![Page 26: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/26.jpg)
Scene Structure Analysis
● Related work: Graph-based symmetry detection (Berner et al. 2008)
– Extract slippage features– Build neighborhood graph– Annotate:
● Edges with their lengths● Features with curvature descriptor
– Find isometric subgraphs
![Page 27: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/27.jpg)
Scene Structure Analysis
● Related work: Graph-based symmetry detection (Berner et al. 2008)
22
1.8
1.52.3
1
1.21.1
1.1
1.2
1.1
1.1
2
![Page 28: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/28.jpg)
Scene Structure Analysis
● Related work: Graph-based symmetry detection (Berner et al. 2008)
22
1.8
1.52.3
1
1.21.1
1.1
1.2
1.1
1.1
2
![Page 29: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/29.jpg)
Scene Structure Analysis
● More approaches:– Mitra et al. 2006, transformation voting
– Pauly et al. 2008, detect grid structures
– ...and lots more
![Page 30: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/30.jpg)
Sub-Symmetry Detection
● My approach: use metric relation of feature points to each other
● Loosely inspired by Berner et al. 2008● Aimed at detecting small sub-symmetries
![Page 31: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/31.jpg)
Sub-Symmetry Detection
● Build neighbor graph over feature points
![Page 32: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/32.jpg)
Sub-Symmetry Detection
● Build neighbor graph over feature points
● For each feature, generate all n-tuples with neighbors
![Page 33: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/33.jpg)
Sub-Symmetry Detection
● Build neighbor graph over feature points
● For each feature, generate all n-tuples with neighbors
● Characterize tuples by a vector of pairwise distances
![Page 34: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/34.jpg)
Sub-Symmetry Detection
● Build neighbor graph over feature points
● For each feature, generate all n-tuples with neighbors
● Characterize tuples by a vector of pairwise distances
● Establish correspondence between similar ones
![Page 35: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/35.jpg)
Sub-Symmetry Detection
● Build neighbor graph over feature points
● For each feature, generate all n-tuples with neighbors
● Characterize tuples by a vector of pairwise distances
● Establish correspondence between similar ones
● “Triangulate” nearby features using tuple features as reference points
● Establish correspondence between similar ones
![Page 36: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/36.jpg)
Sub-Symmetry Detection
● Build neighbor graph over feature points
● For each feature, generate all n-tuples with neighbors
● Characterize tuples by a vector of pairwise distances
● Establish correspondence between similar ones
● “Triangulate” nearby features using tuple features as reference points
● Establish correspondence between similar ones
● Use tuple-relative coordinates to propagate features 2D triangulation
![Page 37: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/37.jpg)
Handling Feature Guesses
● Result: Large set of feature guesses
● Some guesses are wrong● Some are badly constrained● How to handle that robustly?
?
![Page 38: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/38.jpg)
Handling Feature Guesses
● Assign a certainty to each guess
● Merge spatially related guesses
● Boost certainty if many guesses predict the same feature
● Others can be neglected
%
![Page 39: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/39.jpg)
Handling Feature Guesses
● Bonus: Extract feature-to-feature correspondences
● Remember: Guesses were derived from partial symmetries
![Page 40: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/40.jpg)
Overview
● Problem description– Symmetry detection
– Feature extraction
● My approach– The framework
– Scene structure analysis
● Preliminary results
![Page 41: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/41.jpg)
Preliminary Results
● Synthetic example: “IPSFX houses”● 3 iterations
![Page 42: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/42.jpg)
Preliminary Results
No local evidence feature completed
![Page 43: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/43.jpg)
Preliminary Results
● Scanned example: Hannover town hall● Input data and slippage:
![Page 44: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/44.jpg)
Preliminary Results
● Scanned example: Hannover town hall● 1st iteration (interest based features):
![Page 45: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/45.jpg)
Preliminary Results
● Scanned example: Hannover town hall● Propagation quadlets (example):
![Page 46: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/46.jpg)
Preliminary Results
● Scanned example: Hannover town hall● 2nd iteration, features and correspondences:
![Page 47: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/47.jpg)
Preliminary Results
● Scanned example: Hannover town hall● 3rd iteration, features and correspondences:
![Page 48: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/48.jpg)
Preliminary Results
● Scanned example: Hannover town hall● 4th iteration, features and correspondences:
![Page 49: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/49.jpg)
Preliminary Results
● Scanned example: Hannover town hall● 5th iteration, features and correspondences:
![Page 50: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/50.jpg)
Preliminary Results
● Scanned example: Hannover town hall● 6th iteration, features and correspondences:
![Page 51: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/51.jpg)
Preliminary Results
● Scanned example: Hannover town hall● 7th iteration, features and correspondences:
![Page 52: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/52.jpg)
Preliminary Results
Completed window features
![Page 53: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/53.jpg)
Preliminary Results
● Scanned example: Thai elephant statue● Input data and slippage:
![Page 54: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/54.jpg)
Preliminary Results
● Scanned example: Thai elephant statue● 5 iterations
![Page 55: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/55.jpg)
Preliminary Results
“garbage” correspondence class at frequency limit
Correspondences detected at foot
Trunk features nicely completed
![Page 56: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/56.jpg)
Overview
● Problem description– Symmetry detection
– Feature extraction
● My approach– The framework
– Scene structure analysis
● Preliminary results
![Page 57: Extracting Point Features for Symmetry Detectiondanielmewes.dnsalias.net/~daniel/Bachelor Thesis... · Conclusion Good feature extraction requires global understanding of the scene's](https://reader035.vdocuments.mx/reader035/viewer/2022071016/5fcfaa8e46e205613c2ada07/html5/thumbnails/57.jpg)
Conclusion
● Good feature extraction requires global understanding of the scene's structure
● I propose:– A framework for interleaved feature selection
and partial symmetry detection
– A robust approach to sub-symmetry detection
ReferencesGelfand N., Guibas L. J.: Shape segmentation using local slippage analysis. Eurographics 2004Bokeloh M., Berner A., Wand M., Seidel H.-P., Schilling A.: Slippage features. Technical report 2008Berner A., Bokeloh M., Wand M., Schilling A., Seidel H.-P.: A graph based approach to symmetry detection.
Proc. Symp. Point-Based Graphics 2008Pauly M., Mitra N. J., Wallner J., Pottmann H., Guibas L. J.: Discovering Structural Regularity in 3D Geometry.
ACM SIGGRAPH 2008Mitra N. J., Guibas L. J., Pauly M.: Partial and approximate symmetry detection for 3d geometry.
ACM SIGGRAPH 2006