Segmentation Using Population based Markov Chain Monte Carlo

Simulated Annealing is a methodology employed to solve NP-hard problem proximately. Compared with other methods, SA is able to obtain more accurate solution to the problem. However, this algorithm is too costly to be applied to the complicated problems. With this motivation, a novel algorithm Using Population based Markov chain Monte Carlo (Pop-MCMC) is proposed for segmentation. It takes less time from the initial state to the state that chains are coupling with Pop-MCMC than with the Simulated Annealing which is usually employed in traditional MCMC. The main feature Pop-MCMC owns is that multiple samples are generated at a time and information is exchanged between the Markov Chains. A graph is constructed with the atomic regions which are formed using the MeanShift filter. Secondly, the Swendsen-Wang Cuts Algorithm is employed to construct the Markov chain based on the reconstructed energy function. Thirdly, pop-MCMC is employed to speed up the convergence of the Markov chain. Experiments show our algorithm achieves better segmentation results.