Download presentation
Published byJeffry Benson Modified over 9 years ago
1
Fast Approximate Energy Minimization via Graph Cuts
M.S. Student, Hee-Jong Hong May 29, 2013
2
Contents Introduction Previous Works Proposed Method Experiment
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Contents Introduction Previous Works Proposed Method Experiment Conclusion
3
Introduction Local Method Global Method Sum of Squared Differences
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Introduction Local Method Sum of Squared Differences Sum of Absolute Differences Zero-mean Normalized Cross-Correlation Global Method Dynamic Programming (One Dimensional Problem) Graph Cuts (Every Epipolar Line)
4
Introduction Global Optimization V(a,b) = V(b,c) = K/2 V(a,c) = K
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Introduction Global Optimization V(a,b) = V(b,c) = K/2 V(a,c) = K (d) Sum Of Local Energy Sum Of Global Energy (a) 0 + K/2 + K/2 = K (b) 4 = 4
5
Introduction Dynamic Programming 1 2 3 4 Disparity A Image Row
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Introduction Dynamic Programming 1 2 3 4 Disparity A Image Row
6
Introduction Energy Minimization
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Introduction Energy Minimization Another global approach to improve quality of correspondences Assumption: disparities vary (mostly) smoothly Minimize energy function: Edata+lEsmoothness Edata: how well does disparity match data Esmoothness: how well does disparity match that of neighbors – regularization
7
Introduction Energy Definition in Stereo
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Introduction Energy Definition in Stereo
8
Previous Works S T Max Flow / Min Cut A graph with two terminals
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Previous Works Max Flow / Min Cut “source” A graph with two terminals S T “sink”
9
Previous Works Labeling
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Previous Works Labeling For each pixel, either the F or G edge has to be cut Only one edge label per pixel can be cut (otherwise could be added
10
Swap Move & Expansion Move
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Proposed Method Swap Move & Expansion Move
11
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010]
𝜶−𝜷 𝑺𝒘𝒂𝒑
12
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010]
𝜶−𝜷 𝑺𝒘𝒂𝒑
13
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010]
𝜶 − 𝑬𝒙𝒑𝒂𝒏𝒔𝒊𝒐𝒏
14
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010]
𝜶 − 𝑬𝒙𝒑𝒂𝒏𝒔𝒊𝒐𝒏
15
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010]
Move
16
Experiment Energy Definition Data Term :
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Experiment Energy Definition Data Term : Smoothness Term : Static Cues (Weighted Potts)
17
Experiment Static Cues Potts 0?1? unkown Static Cues
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Experiment Static Cues Potts 1Pixel Move 0?1? unkown Static Cues Give Higher Smoothness Factor to Continues Intensity
18
Experiment Expansion Move Swap Move
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Experiment Expansion Move Swap Move
19
Experiment Expansion Move & Swap Move Normalized Corr & Annealling
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Experiment Expansion Move & Swap Move Normalized Corr & Annealling
20
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010]
Experiment
21
Conclusion Performs well on a variety of computer vision problems
Visual Object Tracking using Adaptive Correlation Filters [CVPR 2010] Conclusion Performs well on a variety of computer vision problems Image Restoration, Stereo, and Motion Very Faster than Annealing
22
Thank you!
Similar presentations
© 2025 SlidePlayer.com Inc.
All rights reserved.