Image segmentation via ant colony algorithm and loopy belief propagation algorithm

Loopy belief propagation (LBP) algorithm over Pairwise-connected Markov random fields (MRF) has become widely used for low-level vision problems. However, Pairwise MRFs are often insufficient to capture more expressive priors, and LBP is still extremely slow for application on MRFs with large discrete label space. To solve these problems, a new segmentation algorithm combining ant colony and loopy belief propagation is proposed in this paper. Based on Pairwise MRF, a local interaction region MRF model is constructed. Then ant colony algorithm (ACA) is used to search local optimal label in every local region and to prune the label space for each pixel. Finally the loopy belief propagation algorithm is applied to transfer the local optimal result to adjacent region. This process is iterated until convergence. Compared with some previous algorithms, the proposed algorithm generates more accurate segmentation results and also more speed, because the proposed algorithm utilizes the local optimal result as the propagated messages between nodes in MRF, and uses adaptive label pruning scheme to reduce the number of labels for each pixel, Experimental results on a wide variety of images have verified the effectiveness of the proposed algorithm.