.png){kind=icon}
.png){kind=icon}
Return
Robust image segmentation using EM-based models with TV regularization and alpha-stable distributions
.png){kind=icon}
PRE
AI
S
DOI：10.1007/s00180-026-01730-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper proposes a novel image segmentation framework that integrates alpha-stable mixture models (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha $$\end{document}SMM) with total variation (TV) regularization. Traditional image segmentation methods often rely on Gaussian Mixture Models (GMMs), which assume Gaussian noise characteristics. However, real-world images frequently exhibit impulsive noise or heavy-tailed distributions, rendering Gaussian assumptions inadequate. Our approach leverages the robustness of alpha-stable distributions, which can effectively model such non-Gaussian noise, and combines it with the geometric regularization power of total variation. We develop an Expectation-Maximization (EM) based algorithm to estimate the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha $$\end{document}SMM parameters and incorporate the TV regularization within a unified variational framework. Experimental results on synthetic and real-world images demonstrate the promising performance and robustness of the proposed method, particularly in the presence of impulsive noise, compared to conventional GMM-based approaches.
Keywords:
Image segmentation
Alpha-stable mixture model
Total variation regularization
EM algorithm
Heavy-tailed noise
Impulsive noise
Journal
C
IF:
1.4
Papers:
85
Citations:
2.2K
Organization
Cited Papers