Deterministic Pseudo-Annealing: Optimization in Markow-Random-Fields. an Application to Pixel Classification
Abstract
We present in this paper a new deterministic and massively parallel algorithm for combinatorial optimization in a Markov Random Field. This algorithm is an extension of previous relaxation labeling by optimization algorithms. First, the a posteriori probability of a tentative labeling, defined in terms of a Markov Random Field is generalized to continuous labelings. This merit function of probabilistic vectors is then convexified by changing its domain. Global optimization is performed, and the maximum is tracked down while the original domain is restaured. On an application to contextual pixel quantization, it compares favorably to recent stochastic (simulated annealing) or deterministic (graduated non-convexity) methods popularized for low-level vision.
Cite
Text
Berthod et al. "Deterministic Pseudo-Annealing: Optimization in Markow-Random-Fields. an Application to Pixel Classification." European Conference on Computer Vision, 1992. doi:10.1007/3-540-55426-2_8Markdown
[Berthod et al. "Deterministic Pseudo-Annealing: Optimization in Markow-Random-Fields. an Application to Pixel Classification." European Conference on Computer Vision, 1992.](https://mlanthology.org/eccv/1992/berthod1992eccv-deterministic/) doi:10.1007/3-540-55426-2_8BibTeX
@inproceedings{berthod1992eccv-deterministic,
title = {{Deterministic Pseudo-Annealing: Optimization in Markow-Random-Fields. an Application to Pixel Classification}},
author = {Berthod, Marc and Giraudon, Gérard and Stromboni, Jean Paul},
booktitle = {European Conference on Computer Vision},
year = {1992},
pages = {67-71},
doi = {10.1007/3-540-55426-2_8},
url = {https://mlanthology.org/eccv/1992/berthod1992eccv-deterministic/}
}