Hierarchical GraphCut Phase Unwrapping based on Invariance of Diffeomorphisms Framework

Xiang Gao, Xinmu Wang, Zhou Zhao, Junqi Huang, Xianfeng David Gu · 2025

Read Hierarchical GraphCut Phase Unwrapping based on Invariance of Diffeomorphisms Framework by Xiang Gao, Xinmu Wang, Zhou Zhao, Junqi Huang, Xianfeng David Gu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-18682.txt.

Introduction

Xiang Gao, Xinmu Wang, Zhou Zhao, Junqi Huang, and Xianfeng David Gu unwrap phase maps for structured light 3D scans in a staged graph cut flow. Scanners project striped light and read phase wrapped to a circle; the true phase needs an unknown whole count of turns per pixel. Noise, blocked zones, and bent shapes make that count hard to guess.

Their method leans on an invariance tied to smooth surface warps, framed by diffeomorphisms. A hierarchy of graph cuts then picks the counts with both speed and care, where past tools traded one for the other. Clean phase feeds depth builds for faces and bodies in motion. Uses span virtual scenes, digital humans, and medical imaging.

Catalog record

CREATORXiang Gao, Xinmu Wang, Zhou Zhao, Junqi Huang, Xianfeng David Gu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-18682

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Hierarchical GraphCut Phase Unwrapping based on Invariance of Diffeomorphisms Framework Xiang Gao, Xinmu Wang, Zhou Zhao, Junqi Huang, Xianfeng David Gu 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 2 ​ π 2\pi . The goal is to estimate the unwrapped phase count k k , an integer-valued variable in the equation Φ = ϕ + 2 ​ π ​ k \Phi=\phi+2\pi k , where ϕ \phi is the wrapped phase and Φ \Phi 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 2 ​ π 2\pi 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 k k 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 k k at each pixel, using an odd number of votes to break ties. Experimental results demonstrate a 45.5 × \times speedup and lower L 2 L^{2} error in both real experiments and simulations, showing potential for real-time applications. I Introduction Recent years have seen rapid advancements in 3D scanning technologies, with applications in VR/AR, digital humans, medical imaging, autonomous driving, and robotics. Common approaches include infrared-based systems [ 2 , 1 ] , time-of-flight (ToF) methods [ 5 , 6 ] , and structured-light scanning with phase-shifting techniques [ 15 , 27 , 28 , 25 , 26 ] . While infrared and ToF methods offer real-time capabilities, they suffer from safety concerns or limited spatial accuracy. In contrast, structured-light systems using visible light patterns provide high-resolution and accurate depth maps, making them ideal for human-centric applications such as 4D facial dynamics. However, their performance heavily depends on robust phase unwrapping to convert wrapped fringes into continuous phase maps for depth and 3D point cloud reconstruction. This work addresses a core trade-off in phase unwrapping for structured-light 3D scanning: fast methods tend to produce unwrapping artifacts, while accurate ones are often too slow for real-time use. Phase unwrapping is inherently ill-posed and sensitive to noise, occlusion, and surface reflectance. Existing approaches fall into three main categories: path-following [ 12 , 18 ] , minimum norm [ 11 , 20 ] , and energy-based methods [ 3 , 16 ] . Path-following is efficient but prone to error propagation. Minimum norm methods promote smoothness but may oversimplify sharp features. Energy-based methods like GraphCut [ 3 ] offer high accuracy by minimizing global discontinuity energy, but they are computationally expensive and hard to scale. These challenges point to the need for a method that reduces artifacts while remaining fast and scalable for real-time, high-resolution 3D scanning. To address these challenges, we reinterpret GraphCut-based phase unwrapping as a pixel-labeling problem, forming the foundation of our novel framework. Operating entirely in image space, we leverage invariance under diffeomorphisms by applying conformal mappings and optimal transport (OT) maps to produce conformally equivalent and area-preserving domains. This enables efficient and accurate application of a hierarchical GraphCut algorithm to obtain the unwrapped phase count through majority voting, which is necessary for computing depth and reconstructing the 3D point cloud. Our framework yields substantial gains in computational efficiency and scalability. Experiments show a 45.5× speedup and the lowest L^{2} error among all compared methods. An ablation study further supports its effectiveness in real-world settings, especially near abrupt surface changes where it helps avoid unwrapping artifacts. To summarize, we offer three principal contributions : • To the best of our knowledge, this is the first work to formulate one of the fundamental properties of low-level vision operators: Invariance under Diffeomorphisms (ID) . • We propose a novel phase unwrapping framework to exploit the invariance property, the ID framework, by combining conformal mappings, optimal transport (OT) maps, and ID Hierarchical GraphCut phase unwrapping . • Extensive real-world experiments and simulations demonstrate the effectiveness and robustness of our method, achieving a 45.5× speedup and the lowest L^{2} error compared to state-of-the-art approaches. II Related Works Traditional Phase Unwrapping. Traditional phase unwrapping methods are broadly categorized into path-following, minimum L^{p} -norm, and minimum discontinuity approaches. Path-following methods, such as Goldstein’s algorithm [ 12 ] , use branch cuts to balance residues, while quality-guided methods [ 18 ] improve efficiency by directing unwrapping paths with quality maps, though they remain sensitive to noise. Reliability phase unwrapping [ 21 ] further refines this approach by employing enhanced quality maps. Minimum L^{p} -norm methods, like Poisson solvers for robustness against noise through global phase continuity optimization but suffer from slow convergence. Minimum discontinuity methods, such as Flynn’s algorithm [ 8 ] , segment the wrapped phase into regions and assign integer multiples of 2\pi to minimize discontinuities, excelling where path-following methods fail. Despite their strengths, these traditional approaches struggle to balance computational efficiency with robustness. Energy Based Phase Unwrapping. The use of GraphCut-based methods [ 3 ] significantly advanced phase unwrapping by framing it as an energy minimization problem, enabling higher quality unwrapping results. In many fundamental computer vision tasks, such as image restoration, stereo matching, segmentation, and phase unwrapping, the goal is to assign labels (e.g., intensity, disparity, segmentation regions, or phase counts) to pixels based on noisy measurements. With the presence of uncertainties, the optimal labeling can be determined by minimizing an energy function that reflects the problem’s constraints and objectives. For example, minimum cut/maximum flow algorithms [ 16 , 4 , 19 , 22 ] are widely used for energy minimization, where a graph is constructed such that the minimum cut on the graph corresponds to the optimal energy configuration. The classical use of energy minimization is in solving the pixel labeling problem, where the variables represent individual pixels, and the possible values represent attributes such as displacements or intensities. By leveraging this energy-based framework, GraphCut provides a more accurate and globally consistent phase unwrapping solution, particularly in challenging scenarios with noise and other measurement uncertainties. Pixel Labeling Problem. Suppose \Omega\subset\mathbb{R}^{2} is a planar domain, and a signal I:\Omega\to\mathbb{R} is defined over \Omega . In practice, \Omega is discretized into a set of pixels \mathcal{P} ,

Read the complete text