Back to articles
Proceedings Paper
Volume: 37 | Article ID: COIMG-135
Image
Asymptotically Efficient Simulation and Modeling of Rare Binary Images
  DOI :  10.2352/EI.2025.37.14.COIMG-135  Published OnlineFebruary 2025
Abstract
Abstract

Modeling and simulation of rare events is a problem that has been largely ignored by the image processing community, but is of great interest in many areas of science. Rare events in physical systems are responsible for many failure modes, and as such must be precisely understood. We propose a novel method for simulating and modeling rare images using asymptotically efficient importance sampling, and apply it to binary images of interest in statistical mechanics and materials science. These rare images correspond to the occurrence of rare events in systems modeled by Gibbs distributions, more commonly known as Markov random fields (MRFs) in image processing. We will first give a precise definition of a rare event, in terms of a rare event statistic and a rare event region. The rare event statistic of interest here will be the per-site magnetization of a ferromagnet under the Ising model, but this could be replaced by many other statistics for other problems, such as boundary length in two-phase material microstructures, for example. For the given rare event statistic, we estimate the asymptotically efficient importance sampling (AEIS) distribution, which is based on a large deviation principle (LDP); draw samples of rare binary images from this AEIS distribution; and estimate rare event probabilities for several different rare event regions. Theoretically, the AEIS sampling distribution gives an unbiased estimator with the lowest variance asymptotically from a class of importance sampling distributions that are practically feasible for Monte Carlo Markov chain simulation. Finally, we fit large deviations rate functions from simulations using several rare event regions. This allows us to compute probability estimates associated with a given rare event statistic for any rare event region of interest without requiring further simulations.

Subject Areas :
Views 2
Downloads 1
 articleview.views 2
 articleview.downloads 1
  Cite this article 

Jiaxuan Xu, Mary Comer, "Asymptotically Efficient Simulation and Modeling of Rare Binary Imagesin Electronic Imaging,  2025,  pp 135-1 - 135-7,  https://doi.org/10.2352/EI.2025.37.14.COIMG-135

 Copy citation
  Copyright statement 
Copyright © 2025, Society for Imaging Science and Technology
ei
Electronic Imaging
2470-1173
2470-1173
Society for Imaging Science and Technology
IS&T 7003 Kilworth Lane, Springfield, VA 22151 USA