Hierarchical GraphCut Phase Unwrapping Based on Invariance of Diffeomorphisms Framework

Xiang Gao, Xiaorong Wang, Zhou Zhao, Junqi Huang, Xianfeng GuIEEE Open Journal of Signal Processing 2025 · DOI

TL;DR

This paper introduces a novel phase unwrapping framework for 3D structured-light scanning that leverages the mathematical concept of "invariance under diffeomorphisms." By applying conformal and optimal transport maps to transform image data, they enable a more efficient and accurate hierarchical GraphCut algorithm, achieving significant speedups and reduced errors compared to existing methods. This advancement is crucial for real-time, high-resolution 3D reconstruction, particularly for applications like capturing dynamic facial movements.

What Problem Does This Paper Address?

Structured-light 3D scanning, widely used in fields like VR/AR and digital human creation, relies on phase-shifting techniques to capture detailed surface geometry. A critical step in this process is "phase unwrapping," which converts inherently ambiguous, wrapped phase measurements (modulo 2π) into continuous, absolute phase values. This is essential for accurate depth reconstruction. However, real-world data often suffers from noise, occlusions, and complex surface geometries, making phase unwrapping a challenging, ill-posed problem. Existing methods often face a trade-off: fast algorithms are prone to errors, while accurate ones are too slow for real-time applications. This paper aims to bridge this gap by developing a phase unwrapping method that is both fast and accurate, suitable for high-resolution, real-time 3D scanning.

Methods

The core innovation of this paper lies in reformulating phase unwrapping as a pixel-labeling problem and introducing the concept of "Invariance under Diffeomorphisms" (ID) into the process.

  1. Invariance under Diffeomorphisms (ID) Framework: The authors propose that the underlying geometric continuity or discontinuity of a surface is invariant under certain smooth, invertible transformations of the image space called diffeomorphisms. They leverage this property by applying two specific types of diffeomorphisms:
  2. Conformal Maps (CM): These maps preserve angles and are known to preserve the harmonic energy of a signal, which is related to its smoothness.
  3. Optimal Transport (OT) Maps: These maps are measure-preserving and can be used to "zoom in" on specific regions of interest, enhancing local accuracy.
  4. Deformation and Unwrapping: For each input fringe pattern, they generate multiple transformed versions using these diffeomorphisms. Each transformed image is then processed using a standard phase unwrapping algorithm (specifically, a GraphCut-based method).
  5. Hierarchical GraphCut: The paper utilizes a hierarchical GraphCut approach for phase unwrapping. This involves building a pyramid representation of the image and unwrapping at coarser levels first, using those results as initial guesses for finer levels. This significantly improves computational efficiency.
  6. Majority Voting: The unwrapped phase counts obtained from each of the transformed images are then combined using a majority voting scheme. This robust fusion process helps to mitigate errors introduced by individual transformations or the unwrapping process on those transformed images. An odd number of transformations are used to break ties.

Datasets

The paper evaluates its method using both simulated and real-world data.

  • Simulated Data: A synthetic phase map (double Gaussian) corrupted with noise was used to quantitatively compare the method against traditional algorithms.
  • Real-World Data: Experiments were conducted using a real-time scanning system capturing 3D human face scans. This data was used to compare the proposed method against the baseline GraphCut algorithm.

The paper does not introduce a new benchmark dataset but rather uses existing data acquisition setups and standard evaluation metrics.

Key Insights and Ideas

  • Leveraging Geometric Invariance: The fundamental insight is that the intrinsic properties of a 3D surface (like its smoothness and sharp features) are preserved under certain geometric transformations (diffeomorphisms). This allows for robust unwrapping by applying these transformations and then combining the results.
  • Diffeomorphisms as a Regularizer: Conformal and Optimal Transport maps act as a form of geometric regularization. By transforming the image space, they can simplify challenging regions (e.g., areas with high curvature or noise) for the unwrapping algorithm, making it more robust.
  • Hierarchical GraphCut for Efficiency: The integration of a hierarchical GraphCut approach is key to achieving real-time performance. By processing information at multiple resolutions, it drastically reduces computation time compared to a flat GraphCut.
  • Majority Voting for Robustness: Fusing results from multiple transformed images via majority voting is a simple yet effective way to improve accuracy and reduce the impact of outliers or errors from any single transformation.

Key Results

  • Speedup: The proposed ID Hierarchical GraphCut method achieved a significant speedup of 45.5x compared to the baseline GraphCut algorithm when processing 48 fringe periods.
  • Accuracy: The method demonstrated a lower L² error in both simulated and real-world experiments, outperforming traditional phase unwrapping algorithms, including the standard GraphCut.
  • Artifact Reduction: Qualitative results show that the method effectively corrects unwrapping errors, particularly in regions with sudden surface changes, leading to smoother and more accurate 3D reconstructions.
  • Real-time Potential: The combination of speed and accuracy suggests strong potential for real-time 3D reconstruction applications, such as capturing dynamic facial expressions.

Limitations and Future Work

  • Computational Cost of Diffeomorphisms: While the overall framework is faster, computing the diffeomorphic maps themselves can still be computationally intensive, especially for complex transformations or high-resolution images.
  • Number of Diffeomorphisms: The paper uses N=3 diffeomorphisms. The optimal number and type of transformations might depend on the specific scene and noise levels.
  • Future Work: The authors suggest exploring GPU acceleration for further speed improvements and applying the framework to 4D facial dynamics scanning in real-time. They also implicitly suggest further investigation into the theoretical underpinnings and practical applications of diffeomorphism invariance in computer vision.

Summary updated

Abstract

Recent years have witnessed rapid advancements in 3D scanning technologies, with diverse applications spanning VR/AR, digital human creation, and medical imaging. Structured-light scanning with phase-shifting techniques is preferred for its use of non-radiative, low-intensity visible light and high accuracy, making it well suited for human-centric applications such as capturing 4D facial dynamics. A key step in these systems is phase unwrapping, which recovers continuous phase values from measurements that are inherently wrapped modulo <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$2\pi$</tex-math></inline-formula>. The goal is to estimate the unwrapped phase count <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$k$</tex-math></inline-formula>, an integer-valued variable in the equation <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\Phi = \phi + 2\pi k$</tex-math></inline-formula>, where <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\phi$</tex-math></inline-formula> is the wrapped phase and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\Phi$</tex-math></inline-formula> is the true phase. However, the presence of noise, occlusions, and piecewise continuous phase functions induced by complex 3D surface geometry makes the inverse reconstruction of the true phase extremely challenging. This is because phase unwrapping is an inherently ill-posed problem: measurements only provide modulo <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$2\pi$</tex-math></inline-formula> values, and recovering the correct unwrapped phase count requires strong assumptions about the smoothness or continuity of the underlying 3D surface. Existing methods typically involve a trade-off between speed and accuracy: Fast approaches lack precision, while accurate algorithms are too slow for real-time use. To overcome these limitations, this work proposes a novel phase unwrapping framework that reformulates GraphCut-based unwrapping as a pixel-labeling problem. This framework helps significantly improve the estimation of the unwrapped phase count <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$k$</tex-math></inline-formula> through the invariance property of diffeomorphisms applied in image space via conformal and optimal transport (OT) maps. An odd number of diffeomorphisms are precomputed from the input phase data, and a hierarchical GraphCut algorithm is applied in each corresponding domain. The resulting label maps are fused via majority voting to efficiently and robustly estimate the unwrapped phase count <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$k$</tex-math></inline-formula> at each pixel, using an odd number of votes to break ties. Experimental results demonstrate a 45.5× speedup and lower <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$L^{2}$</tex-math></inline-formula> error in both real experiments and simulations, showing potential for real-time applications.
BibTeX
@article{gao2025hierarchical,
  author = {Xiang Gao and Xiaorong Wang and Zhou Zhao and Junqi Huang and Xianfeng Gu},
  title = {{Hierarchical GraphCut Phase Unwrapping Based on Invariance of Diffeomorphisms Framework}},
  journal = {IEEE Open Journal of Signal Processing},
  year = {2025},
  doi = {10.1109/ojsp.2025.3568757},
}
Figure