Issue No. 05 - May (2002 vol. 24)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/34.1000239
<p>This paper presents a computational paradigm called <it>Data-Driven Markov Chain Monte Carlo (DDMCMC)</it> for image segmentation in the Bayesian statistical framework. The paper contributes to image segmentation in four aspects. First, it designs efficient and well-balanced Markov Chain dynamics to explore the complex solution space and, thus, achieves a nearly global optimal solution independent of initial segmentations. Second, it presents a mathematical principle and a <it>K-adventurers</it> algorithm for computing multiple distinct solutions from the Markov chain sequence and, thus, it incorporates intrinsic ambiguities in image segmentation. Third, it utilizes data-driven (bottom-up) techniques, such as clustering and edge detection, to compute importance proposal probabilities, which drive the Markov chain dynamics and achieve tremendous speedup in comparison to the traditional jump-diffusion methods. Fourth, the DDMCMC paradigm provides a unifying framework in which the role of many existing segmentation algorithms, such as, edge detection, clustering, region growing, split-merge, snake/balloon, and region competition, are revealed as either realizing Markov chain dynamics or computing importance proposal probabilities. Thus, the DDMCMC paradigm combines and generalizes these segmentation methods in a principled way. The DDMCMC paradigm adopts seven parametric and nonparametric image models for intensity and color at various regions. We test the DDMCMC paradigm extensively on both color and gray-level images and some results are reported in this paper.</p>
Image segmentation, Markov Chain Monte Carlo, region competition, data clustering, edge detection, Markov random field
Z. Tu and S. Zhu, "Image Segmentation by Data-Driven Markov Chain Monte Carlo," in IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 24, no. , pp. 657-673, 2002.