Fast Unsupervised Bayesian Image Segmentation With Adaptive Spatial Regularisation.

Area of Science:

• Computer Vision
• Statistical Modeling
• Image Processing

Background:

• Hidden Potts-Markov random fields are complex models for image analysis.
• Estimating regularization parameters in these models is computationally challenging.
• Unsupervised image segmentation requires robust and efficient algorithms.

Purpose of the Study:

• To develop a novel Bayesian estimation technique for hidden Potts-Markov random fields.
• To enable fast and unsupervised K-class image segmentation.
• To automatically adapt regularization parameters during the segmentation process.

Main Methods:

• Marginalization of the regularization parameter from the Bayesian model.
• Small-variance-asymptotic (SVA) analysis to decouple Potts model terms.
• Iterative solution combining convex total-variation denoising and K-means clustering.
• Application of parallel computing for high-dimensional data.

Main Results:

• The proposed method achieves extremely fast convergence.
• Accurate image segmentation results are obtained on synthetic and real data.
• The methodology demonstrates self-adjusting regularization parameters.
• Effective application in large 2D and 3D scenarios is confirmed.

Conclusions:

• The developed Bayesian estimation technique offers a fast and fully unsupervised image segmentation solution.
• The method's ability to automatically adapt regularization parameters enhances its practical utility.
• The approach is computationally efficient and scalable for complex image analysis tasks.