Enhanced fringe-to-phase framework using deep learning

TL;DR

SFNet tackles the need for fast, robust 3D surface measurement with minimal fringe data by learning to predict the absolute phase from two fringe images. It uses a symmetric dual-encoder, dual-decoder architecture to predict the numerator and denominator of the wrapped phase, which are then combined into the absolute phase, and it enhances reliability through refined reference phases derived from multiple frequencies during training. A minimum-phase based and refined wrapped phase strategy provides high-quality references, enabling performance close to multi-frequency TPU with just two inputs. Validated on a large synthetic SynthFringe dataset, SFNet achieves MAE about 0.0527 rad and RMSE about 0.654 rad, outperforming several baselines, demonstrating a practical path toward real-time FPP with few patterns.

Abstract

In Fringe Projection Profilometry (FPP), achieving robust and accurate 3D reconstruction with a limited number of fringe patterns remains a challenge in structured light 3D imaging. Conventional methods require a set of fringe images, but using only one or two patterns complicates phase recovery and unwrapping. In this study, we introduce SFNet, a symmetric fusion network that transforms two fringe images into an absolute phase. To enhance output reliability, Our framework predicts refined phases by incorporating information from fringe images of a different frequency than those used as input. This allows us to achieve high accuracy with just two images. Comparative experiments and ablation studies validate the effectiveness of our proposed method. The dataset and code are publicly accessible on our project page https://wonhoe-kim.github.io/SFNet.

Figures (13)

• Figure 1: Framework of 3D reconstruction using the conventional FPP-based technique.
• Figure 2: Illustration of the proposed method. The neural network structure of the red-dashed box (a) is schematized in (b).
• Figure 3: Frameworks of MF-TPU method and modified MF-TPU. To implement the TPU method, a reference phase without phase ambiguity is essential. (a) In the conventional MF-TPU, a unit frequency fringe pattern is employed for this purpose. (b) The modified approach constructs a minimum phase using geometric constraints, thereby eliminating phase ambiguity.
• Figure 4: Framework of obtaining the refined reference phase.
• Figure 5: Comparison of the two reference phases obtained in Fig. \ref{['fig:tpuvszmin']}. (a) 3D plot and phase difference between the three-frequency TPU method shown in Fig. \ref{['fig:3ftpumethod']}) and minimum-phase method shown in Fig. \ref{['fig:zminmethod']}. (b) Cross-lines of the reference phase, fringe order, and unwrapped phase are marked in the plain image in (a).