Near Field Illumination Modeling for Inverse Rendering

16 concepts16 papers1 proposal

In classical computer vision, researchers often rely on the distant illumination assumption. This simplification treats light sources as if they are infinitely far away, resulting in light rays that are perfectly parallel and uniform in intensity across an entire scene. While mathematically convenient, this assumption collapses in the real world—especially in indoor environments, handheld scanning, or macro photography—where light sources are proximal to the objects they illuminate.

Near-field illumination modeling replaces these idealized parallel rays with the messy, spatially varying physics of local light sources. By accounting for how light diverges and weakens as it travels, these methods enable significantly more accurate 3D reconstruction and material estimation.

The Geometry of Proximity

The most immediate consequence of a near-field light source is the breakdown of directional uniformity. When a light source is far away, every point on an object "sees" the light coming from the same direction. However, as Hara et al. et al. demonstrate, proximal light sources cause significant variations in incidence angles across a surface.

This geometric shift is critical for techniques like Photometric Stereo (PS), which recovers shape by observing how shadows and highlights move under different lighting. Sato et al. et al. showed that even the shadows cast by an object can be used to map these complex, non-parallel light distributions. If a system assumes parallel rays when the rays are actually diverging from a nearby point source, the resulting 3D model will suffer from severe geometric distortions.

The Inverse Square Law

Beyond direction, near-field modeling must account for attenuation. According to the inverse-square law, the intensity of light decreases quadratically with the distance from the source. In a far-field setup, this effect is negligible because the distance from the light to the object is effectively constant.

In near-field scenarios, such as scanning a foot sole on a flatbed scanner (Martedi et al. et al.) or using a smartphone flash to capture facial details (Wang et al. et al.), the distance from the light to different parts of the object varies significantly. Lichy et al. et al. highlight that failing to model this attenuation leads to massive errors in surface normal estimation, as the system might mistake a drop in light intensity for a change in surface orientation.

Modeling Point Sources and Local Reflectors

To handle these complexities, researchers have moved toward more sophisticated mathematical frameworks:

  • Point Light Models: Instead of a single global light vector, Durix et al. et al. and Logothetis et al. et al. model light sources (like LEDs) as point sources. This requires solving systems of partial differential equations (PDEs) that explicitly include the $1/r^2$ distance term.
  • Neural Plenoptic Functions: Modern inverse rendering often uses neural fields to represent light. Zarzar et al. et al. and the authors of Inverse Rendering of Glossy Objects via the Neural Plenoptic Function and Radiance Fields argue that 2D environment maps are insufficient for glossy objects. They instead model illumination as a 5D function that accounts for light reflecting off nearby surfaces, capturing the "local" environment rather than just distant stars or studio lights.

Why Near-Field Modeling Works

The shift from far-field to near-field modeling works because it increases the physical fidelity of the inverse rendering pipeline. By acknowledging that light intensity and direction are functions of 3D position—rather than constants—these models can disambiguate between light effects and geometric features.

For instance, Santo et al. et al. show that by modeling the spatial variance of light, it is possible to reconstruct surfaces with complex, spatially varying reflectances that would otherwise confuse a far-field system. Similarly, Chan et al. et al. use reinforcement learning to plan light positions that maximize the information gained from these near-field effects, proving that the complexity of proximal lighting is not just a hurdle to be overcome, but a rich source of geometric information.

Key Foundational Work

Illumination Distribution from Shadows (1999) — contributed 1 concept elements to this direction.
Determining Reflectance and Light Position from a Single Image Without Distant Illumination Assumption (2003) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_near_field_illumination_modeling_ed886eed,
  title = {near-field illumination modeling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/near-field-illumination-modeling/rdr_ed886eed24c94b60a0e4326b8cd52772},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_near_field_illumination_modeling_ed886eed_v1,
  title = {near-field illumination modeling},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/near-field-illumination-modeling/rdr_ed886eed24c94b60a0e4326b8cd52772/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake