Phase unwrapping via graph cuts

Area of Science:

• Computer Vision
• Image Processing
• Optimization

Background:

• Phase unwrapping is crucial for reconstructing absolute phase from wrapped phase data.
• Existing methods face challenges with complex phase data and computational efficiency.
• First-order Markov random fields are commonly used for phase unwrapping objective functions.

Purpose of the Study:

• To introduce a new energy minimization framework for phase unwrapping.
• To develop efficient algorithms for both convex and non-convex objective functions.
• To demonstrate the effectiveness and competitiveness of the proposed approach.

Main Methods:

• Developed an energy minimization framework based on first-order Markov random fields.
• Proposed an exact energy minimization algorithm for convex clique potentials using max-flow/min-cut.
• Devised an approximate solution for NP-hard non-convex clique potentials using graph cut techniques.
• Named the algorithms PUMA (phase unwrapping max-flow/min-cut).

Main Results:

• Achieved an exact energy minimization algorithm with complexity KT(n, 3n) for convex potentials.
• Provided an effective approximate solution for non-convex potentials, addressing NP-hard problems.
• Experimental results show PUMA's effectiveness and competitiveness against state-of-the-art methods.

Conclusions:

• The PUMA framework offers a robust and efficient approach to phase unwrapping.
• The method demonstrates strong performance, particularly for challenging phase data.
• PUMA presents a competitive alternative to existing phase unwrapping algorithms.