Texturing Internal Surfaces from a Few Cross Sections

Nico Pietroni, Miguel A. Otaduy, Bernd Bickel, Fabio Ganovelli, Markus H. GrossComputer Graphics Forum 2007 · DOI

TL;DR

This paper introduces a novel paradigm for synthesizing realistic internal textures of 3D models using just a few cross-section photographs of a real object. Their approach leverages a BSP tree-based interpolation domain and an efficient texture morphing algorithm to generate textures on-demand, enabling interactive applications like digital sculpting and fracture simulations.

What Problem Does This Paper Address?

Computer graphics has long used textures to add visual detail to 3D models, primarily for external surfaces. However, when objects are cut, carved, or fractured, their internal structures are revealed, and texturing these internal surfaces realistically has been a significant challenge. Existing methods are often limited in scope, require extensive manual input, or fail to produce photorealistic results. The authors aim to bridge this gap by enabling users to create compelling internal textures from a minimal set of real-world examples, making it easier to model complex organic objects and simulate dynamic processes.

Methods

The core of their contribution lies in a two-pronged approach:

  • BSP Tree Interpolation Domain: They propose using a Binary Space Partitioning (BSP) tree to represent the interior of a 3D model. This structure naturally mimics how a physical object is divided by planar cuts. Each region within the BSP is associated with a subset of input exemplars, simplifying the interpolation of appearance properties.
  • Efficient Texture Morphing Algorithm: They developed a novel texture morphing algorithm that synthesizes texture attributes on a per-point, on-demand basis. This is crucial because, unlike full texture synthesis, each point on an internal surface typically requires a unique combination of source points and weights from the exemplars. Their algorithm is inspired by existing warping-based morphing techniques but is optimized for this specific on-demand scenario. Key aspects include:
  • Curved Projections and Scattered Data Interpolation: Used to determine source points and their weights within the BSP regions.
  • Approximation of Inverse Warping: A computationally efficient method to quickly determine where a target point maps back to the exemplar images, avoiding the high cost of exact inverse warping.
  • Local Histogram Matching: A feature enhancement technique adapted for on-demand computation, preserving small-scale details that might otherwise be blurred during morphing.

Datasets

Not discussed in the paper. The authors primarily use photographs of real objects (kiwi, orange, onion, cabbage) as input exemplars and demonstrate their method on various 3D models (bunny, dragon, organic shapes).

Key Insights and Ideas

  • BSP Trees as a Natural Fit: The paper highlights that BSP trees are an intuitive and effective data structure for representing the spatial partitioning of an object's interior, mirroring real-world cutting processes.
  • On-Demand Synthesis is Key: For internal surfaces, where each point's texture contribution is unique, an on-demand synthesis approach is far more efficient and scalable than generating a full volumetric texture.
  • Approximation for Efficiency: The development of an efficient approximation for inverse warping is a critical technical innovation that enables real-time performance for per-point texture synthesis.
  • Local Feature Enhancement: Adapting histogram matching to a local kernel around the target point is a clever way to retain fine details without the computational overhead of global texture processing.

Key Results

  • Real-time Performance: The implemented system achieves a throughput of approximately 20,000 points/sec on a laptop, making interactive applications feasible.
  • Photorealistic Internal Textures: The method successfully generates highly realistic internal surfaces, even for complex organic shapes and when morphing between dissimilar textures.
  • Versatile Applications: Demonstrated effectiveness in carving 3D models, interactive cutting/fracture simulations, and transferring textures between different objects.
  • Robustness to Input: The algorithm handles cases where exemplar colors don't perfectly match at intersections and can work with partial samples or even synthetic exemplars.

Limitations and Future Work

  • Blurring of Fine Features: While local histogram matching helps, the authors acknowledge that morphing can still lead to some blurring of very fine features, especially when exemplars are significantly different.
  • Limitations of Pure Morphing: For highly dissimilar exemplars, a purely morphing-based approach might not be sufficient to avoid all blur, suggesting future work could explore combinations with stereology or pattern-matching optimization.
  • Color Synthesis Only: The current implementation focuses on color synthesis. Future work could extend the framework to other appearance attributes like bump mapping, though this might be challenging due to the global feature-capturing strength of their morphing approach.
  • Parallel Implementation: The authors suggest exploring parallel implementations on graphics hardware to further accelerate the morphing stage, potentially by simplifying the BSP tree to a K-d tree for axis-aligned cross-sections.

Summary updated

BibTeX
@article{pietroni2007texturing,
  author = {Nico Pietroni and Miguel A. Otaduy and Bernd Bickel and Fabio Ganovelli and Markus H. Gross},
  title = {{Texturing Internal Surfaces from a Few Cross Sections}},
  journal = {Computer Graphics Forum},
  year = {2007},
  doi = {10.1111/j.1467-8659.2007.01087.x},
}
Figure