Unsupervised-clustering-driven noise-residue filter for phase images.

Area of Science:

• Optical Metrology
• Image Processing
• Computational Science

Background:

• Phase inconsistencies, or residues, in wrapped phase maps are a primary source of phase error propagation.
• Effective filtering of these residues is crucial for accurate two-dimensional (2D) phase unwrapping.

Purpose of the Study:

• To propose and validate an unsupervised-clustering-driven filter for noise-residue reduction.
• To enhance the preprocessing stage of 2D phase unwrapping algorithms.

Main Methods:

• Developed an unsupervised clustering filter to identify and remove adjacent residues caused by noisy wrapped phases.
• Implemented an adaptive local window size determination based on noise levels.
• Converted path-following unwrapping to path independence by eliminating the need for branch cuts.

Main Results:

• Demonstrated significant reduction in noise-induced residues.
• Confirmed effective preservation of fringe patterns.
• Achieved substantial improvements in the operating speed of phase unwrapping.

Conclusions:

• The proposed unsupervised clustering filter is a valid and efficient preprocessing step for 2D phase unwrapping.
• The method offers improved accuracy, fringe preservation, and processing speed for optical measurements.