High-contrast computational caustic design

Yuliy Schwartzburg, Romain Testuz, Andrea Tagliasacchi, Mark PaulyACM Trans. Graphics 2014 · DOI

TL;DR

This paper introduces a novel computational approach for designing transparent objects that precisely sculpt light into intricate, high-contrast caustic patterns. By leveraging optimal transport and adaptive discretization, their method overcomes limitations of prior work, enabling the creation of caustics with sharp details, completely black areas, and even infinite light density singularities, opening new possibilities for design and art.

What Problem Does This Paper Address?

Creating specific, detailed light patterns (caustics) using refractive objects is notoriously difficult. Traditional methods of shaping materials are challenging because small changes to a surface can have large, unpredictable effects on the resulting light patterns. While previous computational methods have attempted to solve this "inverse problem" (designing an object to produce a desired light pattern), they often struggle with high-contrast images, sharp details, and completely dark regions. These limitations restrict the aesthetic and functional possibilities of caustic design, particularly for applications requiring intricate patterns or dramatic lighting effects.

Methods

The core of this paper's contribution lies in its novel computational framework for inverse caustic design. They frame the problem as an optimal transport task.

  • Optimal Transport Formulation: Instead of directly optimizing the surface shape, they first compute an "optimal transport map" (OTM). This map defines how light rays from the source distribution on the receiver screen should be moved to match the desired target caustic image. This is a powerful technique because it naturally handles the redistribution of light and can accommodate discontinuities.
  • Adaptive Discretization: To handle the complex target distributions, especially those with sharp transitions and singularities, they employ an adaptive discretization scheme. This allows the computational mesh to refine itself in areas of high detail, ensuring accurate representation of the target.
  • Singularity Support: A key innovation is the explicit support for point and curve singularities of infinite light density in the target caustic image. This is achieved by defining base functions for these singularities and integrating them into the optimal transport formulation.
  • 3D Optimization: After computing the OTM, a subsequent iterative 3D optimization process refines the surface of the "caustic generator" (the transparent object). This optimization ensures that the generated surface normals align with the target normals derived from the OTM, while also preserving light flux and maintaining well-shaped geometry. Crucially, their method allows for discontinuities in the surface normals, which is essential for creating high-contrast caustics.

Datasets

Not discussed in the paper. The paper focuses on generating target caustic images from user-defined inputs (e.g., SVG files for shapes, PNGs for textures) and then designing the physical object.

Key Insights and Ideas

  • Optimal Transport as a Natural Fit: Optimal transport is exceptionally well-suited for this problem because it inherently deals with mapping distributions and can gracefully handle the discontinuities and singularities that are characteristic of high-contrast caustics.
  • Adaptive Discretization is Crucial: The ability to adapt the computational mesh to the complexity of the target image is what allows the algorithm to accurately capture fine details and sharp transitions, which were major limitations in prior work.
  • Singularities Unlock New Aesthetics: Explicitly modeling and supporting infinite light density singularities (points and curves) dramatically expands the design space, enabling visually striking effects that were previously impossible.
  • Discontinuities are a Feature, Not a Bug: Unlike methods that enforce surface smoothness everywhere, this approach embraces discontinuities in surface normals. This is fundamental for creating areas of complete darkness within a caustic pattern.

Key Results

The paper demonstrates impressive results across several examples:

  • High-Contrast Caustics: They successfully generate caustics with sharp, detailed patterns, including complex shapes like a human brain and the Olympic rings, with completely black areas between the light features.
  • Singularity Reproduction: They show examples where light is focused into sharp points and curves, mimicking infinite light density.
  • Physical Prototypes: Fabricated acrylic prototypes closely match the simulated results, validating the effectiveness of their design approach. For instance, the "Brain" example shows a physically realized object producing a high-contrast caustic that matches the simulation.
  • Comparison to State-of-the-Art: Figures 3 and 12 show that their method outperforms previous leading techniques (Poisson-based continuous and brightness warping methods) in achieving high contrast and avoiding artifacts, especially on dark backgrounds. While previous methods require artificially brightening backgrounds to reduce artifacts, this method handles them directly.
  • Performance: While computationally intensive (e.g., 4 hours for a large test case), the multi-scale optimal transport approach offers a reasonable trade-off for achieving high-quality results.

Limitations and Future Work

  • Computational Cost: The optimal transport computation, particularly power diagram recomputation, is the bottleneck. While multi-scale strategies help, it's still more computationally expensive than simpler methods for low-contrast images.
  • Idealized Assumptions: The model assumes a single incident light direction (idealized illumination like parallel light or point sources) and perfect specular scattering. Real-world scenarios with area lights or diffuse illumination will introduce blur.
  • Fabrication Imperfections: Physical prototypes show slight distortions due to machining and polishing inaccuracies, though the overall quality is maintained.
  • Future Work:
    • Investigate more sophisticated regularization terms to prevent triangle inversions during optimization.
  • Explore handling area light sources and their impact on caustics.
  • Consider the wave nature of light and diffraction for even finer details.
  • Extend the framework to handle multiple source surfaces, dynamic caustics, or optimizing for multiple caustic images simultaneously.
  • Apply the performative form-finding approach to other domains like geometry editing, digital fabrication, and physical simulation.

Summary updated

BibTeX
@article{schwartzburg2014high,
  author = {Yuliy Schwartzburg and Romain Testuz and Andrea Tagliasacchi and Mark Pauly},
  title = {{High-contrast computational caustic design}},
  journal = {ACM Trans. Graphics},
  year = {2014},
  doi = {10.1145/2601097.2601200},
}
Figure