Markov Random Fields for Edge Classification
description
Transcript of Markov Random Fields for Edge Classification
![Page 1: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/1.jpg)
Markov Random Fields for Edge Classification
Grant Schindler
CS 7636 Final Project
![Page 2: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/2.jpg)
Problem: Edge Classification
Given vanishing points of a scene, classify each pixel according to vanishing direction
![Page 3: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/3.jpg)
MAP Edge Classifications
Red: VP1 Green: VP2 Blue: VP3 Gray: Other White: Off
![Page 4: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/4.jpg)
Bayesian Model
p(M | G,V) = p(G | M,V) p(M) / ZM = classifications, G = gradient magnitude/direction, V = vanishing points
Prior: p(m)
Likelihood: p(g | m,V)
Independent Prior MRF Prior
m
g
![Page 5: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/5.jpg)
Classifications w/MRF Prior
Gibbs sampling over 4-neighbor lattice w/ clique potentials defined as: A if i=j, B if i <> j
![Page 6: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/6.jpg)
Gibbs Sampling & MRFs
Gibbs sampling approximates posterior distribution over classifications at each site (by iterating and accumulating statistics)
Sample from distribution over labels for one site conditioned on all other sites in its Markov blanket
![Page 7: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/7.jpg)
Directional MRF
Give more weight to potentials of neighbors which lie along the vanishing direction of current model
vp
![Page 8: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/8.jpg)
Original Image
![Page 9: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/9.jpg)
Independent Prior
![Page 10: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/10.jpg)
MRF Prior
![Page 11: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/11.jpg)
Directional MRF Prior
![Page 12: Markov Random Fields for Edge Classification](https://reader034.fdocuments.net/reader034/viewer/2022051316/568149d5550346895db6fd63/html5/thumbnails/12.jpg)
Conclusion, Discussion
• This is really the E-Step of an EM process that alternates between edge classification and vanishing point estimation - will MRFs improve the vanishing point estimates?
• Questions?