Stereo Matching Using Population-Based MCMC

Joonyoung Park, Wonsik Kim, Kyoung Mu Lee

Abstract: In this paper, we propose a new stereo matching method using the population-based Markov Chain Monte Carlo (Pop-MCMC). Pop-MCMC belongs to the sampling-based methods. Since previous MCMC methods produce only one sample at a time, only local moves are available. However, since Pop-MCMC uses multiple chains and produces multiple samples at a time, it enables global moves by exchanging information between samples, and in turn leads to faster mixing rate. In the view of optimization, it means that we can reach a state with the lower energy. The experimental results on real stereo images demonstrate that the performance of proposed algorithm is superior to those of previous algorithms.