Unsupervised statistical segmentation of nonstationary images using triplet Markov fields.

Area of Science:

• Statistical modeling
• Image processing
• Computer vision

Background:

• Hidden Markov Field (HMF) models are popular for image segmentation due to their simplicity and effectiveness in stationary cases.
• However, HMF models often fail when dealing with nonstationary image data.
• There is a need for advanced models that can handle nonstationary fields and complex noise characteristics.

Purpose of the Study:

• To propose an original approach for modeling nonstationary hidden random fields for unsupervised statistical image segmentation.
• To address the limitations of existing models in handling nonstationary image data.
• To develop a method capable of dealing with correlated and non-Gaussian noise that varies with image class.

Main Methods:

• Utilized the Triplet Markov Field (TMF) model to represent nonstationary class fields.
• Developed an original parameter estimation method employing the Pearson system to characterize noise distributions.
• Applied the TMF model and parameter estimation for unsupervised image segmentation.

Main Results:

• The proposed TMF model effectively handles nonstationary hidden random fields.
• The parameter estimation method accurately identifies varying noise characteristics across different classes.
• Experimental results demonstrate improved segmentation performance compared to classical methods.

Conclusions:

• The Triplet Markov Field (TMF) model offers a significant advancement for unsupervised statistical image segmentation in nonstationary scenarios.
• The developed parameter estimation technique enhances the model's robustness to complex noise.
• This approach provides improved image quality and segmentation accuracy for challenging image datasets.