Disparity Estimation by Graph Cut/Belief Propagation
for Rectified Image Pair(Triple)
Disparity estimation is an image labeling problem. It is modeled by Markov Random Field (MRF), and the energy minimization task is solved by some popular global optimization methods, i.e. Graph Cut and Belief Propagation. If we assume the image pair (triple) comes from rectified stereo, then the disparity estimation turns to be a 1-d searching process. Below we give some results by applying a Graph Cut-based method proposed by Boykov, Veksler and Zabih (BVZ), its source code can be downloaded from Kolmogorov's web page. Meanwhile, some respective results are given by applying Belief Propagation-based method proposed by Felzenszwalb and Huttenlocher (FH), its source code can be downloaded from Felzenszwalb's web page. More comparison and evaluation of various stereo matching algorithms and source codes can be seen at the Middlebury Stereo Vision Page.
Graph Cut-based Stereo matching (BVZ)
"Tsukuba".
"Rena".
"Akko&Kayo".
Belief Propagation-based Stereo matching (FH)
"Tsukuba".
"Rena".
"Akko&Kayo".