Abstract. A mathematical model for fibrous structures using a direction dependent scaling law is presented. The orientation of fibrous nets (e.g. paper) is analysed with a method based on the curvelet transform. The curvelet-based orientation analysis has been tested successfully on real data from paper samples: the major directions of fibre orientation can apparently be recovered. Similar results are achieved in tests on data simulated by the new model, allowing a comparison with ground truth.
1. Introduction
In industries such as papermaking, the quality of the final product is heavily dependent on the distribution and orientation of wood fibers within a planar network. Manufacturers typically use optical transmission images to monitor these properties in real-time. However, these images are notoriously noisy due to light scattering within the fibers, making precise calibration difficult.
To address this, researchers require a robust framework that combines high-fidelity simulation with advanced analytical tools. This paper introduces a dual-pronged approach: the "H model" for simulating realistic fiber networks where the "ground truth" is known, and a curvelet-transform-based method for precisely identifying fiber orientation. By integrating these tools, it becomes possible to calibrate optical measurements against known physical parameters.
2. Analysis via Curvelet Transforms
Traditional analytical tools like the Fourier transform or standard wavelets often struggle to capture highly directional, curved features. This study utilizes the curvelet transform, which is specifically designed to be "tightly localized" in both space and frequency.
Curvelets follow a parabolic scaling law, making them act as specialized sensors for directional features. As the scale $a$ decreases (zooming in), the curvelets become increasingly needle-like. This property makes them mathematically optimal for detecting thin, linear structures like wood fibers. By analyzing the coefficients of the curvelet transform, researchers can determine if a fiber exists at a specific location with a specific orientation ($\theta$).
3. Simulation via the "H model"
To validate the analysis tool, the authors propose the H model for creating synthetic fiber networks. This model treats the network as a stationary Gaussian random field based on fractional Brownian motion statistics.
The key parameter is the Hurst index ($H$), which controls the "roughness" or feature size of the field. By assigning different Hurst indices to different directions, the model can simulate anisotropic networks where fibers exhibit a preferred orientation. This mimics the scaling laws observed in real physical structures, providing a controllable environment for testing.
4. Results and Validation
The proposed methods were evaluated using both real paper samples and simulated datasets. The curvelet-based analysis successfully recovered major fiber directions in physical materials, matching previous experimental benchmarks.
In simulated tests, the method demonstrated high precision. When a simulated net was rotated by 30°, the resulting orientation curve shifted by the exact same amount. The method also remained effective under shearing distortions, although such transformations affected the relative strength of the orientation peaks. These results confirm that the H model produces the desired orientations and serves as a reliable "sandbox" for testing future image-processing algorithms.
5. Conclusion
The H model and curvelet analysis provide a powerful framework for understanding fibrous materials. While the current model primarily focuses on perpendicular orientations common in newsprint, future work will aim to simulate more complex, multi-directional distributions. The ultimate goal is an "inverse" workflow: analyzing a real paper scan to determine its orientation distribution and then using the H model to generate a synthetic twin for deeper structural study.
Summary updated