Related Experiment Video
Updated: Jul 7, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Unsupervised image restoration and edge location using compound Gauss-Markov random fields and the MDL principle.
1Dept. de Engenharia Electrotecnica e de Comput., Inst. Superior Tecnico, Lisbon.
Summary
This study introduces a new unsupervised method for image restoration that treats edges as unknown parameters. It uses the minimum description length (MDL) principle to find optimal edge locations for clearer restored images.
Area of Science:
- Computer Vision
- Image Processing
- Statistical Modeling
Background:
- Bayesian image restoration often uses two Markov random fields for intensities and edges.
- Joint maximum a posteriori (MAP) estimation requires specifying priors for both fields.
Purpose of the Study:
- To develop a new unsupervised discontinuity-preserving image restoration criterion.
- To infer discontinuity locations directly from image data without additional assumptions.
Main Methods:
- Interpreting discontinuity locations as deterministic unknown parameters within a compound Gauss-Markov random field (CGMRF).
- Employing the minimum description length (MDL) principle to determine the optimal number and configuration of edges.
- Treating noise and CGMRF variances as unknown parameters.
Main Results:
- A novel unsupervised criterion for discontinuity-preserving image restoration is proposed.
- A continuation-type iterative algorithm is developed for implementation.
- The algorithm estimates the number and locations of discontinuities, noise variance, image variance, and the restored image.
Conclusions:
- The proposed method effectively restores images while preserving discontinuities without requiring edge priors.
- The MDL principle provides a robust way to handle an unknown number of edge parameters.
- Experimental results demonstrate the efficacy of the approach on real and synthetic images.
