Estimating Gaussian Markov random field parameters in a nonstationary framework: application to remote sensing...

Area of Science:

• Image analysis and computer vision
• Statistical modeling
• Remote sensing

Background:

• Accurate estimation of textural parameters is crucial for image segmentation.
• Nonstationarities in local mean can significantly challenge texture analysis.
• Existing methods often struggle with variations in local image properties.

Purpose of the Study:

• To develop and evaluate novel methods for estimating textural parameters in nonstationary image data.
• To improve texture feature extraction for image segmentation tasks.
• To apply these methods to real-world remote sensing data for practical applications.

Main Methods:

• Proposed two estimation methods for Gaussian Markov random fields in a nonstationary framework.
• Method 1: Conditional probability extraction and least square approximation for piecewise constant local mean.
• Method 2: Renormalization theory for variance estimation, yielding Cramer-Rao estimators.

Main Results:

• Both methods effectively estimate textural parameters in nonstationary conditions.
• Method 1 reduces blurring at texture edges, enhancing discrimination in single images.
• Method 2 demonstrates robustness to sampling size and local mean nonstationarities.

Conclusions:

• The developed textural parameter estimation methods enable accurate texture discrimination in remote sensing.
• Application to SPOT and AVHRR data successfully delineated urban areas and segmented ice regions.
• The proposed techniques offer significant improvements for image segmentation tasks involving complex textural variations.