Recovery of polarimetric Stokes images by spatial mixture models

Area of Science:

• Image processing
• Computational physics
• Computer vision

Background:

• Polarization-encoded images offer rich information but require advanced processing for accurate analysis.
• Existing methods may struggle with physical constraints and solution smoothness in polarization image analysis.

Purpose of the Study:

• To develop a novel Bayesian framework for joint restoration and segmentation of polarization-encoded images.
• To ensure solutions are both physically admissible and spatially smooth.

Main Methods:

• A Bayesian approach is employed for simultaneous image restoration and segmentation.
• Two models are utilized: one based on Stokes vectors and truncated Gaussian mixtures, and another using coherency matrices parameterized by Gaussian mixtures.
• Physical admissibility is enforced by assigning zero probability to invalid configurations.

Main Results:

• The proposed Bayesian method successfully restores and segments polarization-encoded images.
• Both Stokes vector and coherency matrix models demonstrate effectiveness.
• The approach yields physically plausible and smooth image representations.

Conclusions:

• The developed Bayesian framework provides a robust method for analyzing polarization-encoded images.
• The emphasis on physical admissibility and solution smoothness enhances the reliability of image restoration and segmentation.