Near-Light Color Photometric Stereo for Mono-Chromatic Non-Lambertian Surfaces

Zonglin Li, Jieji Ren, Shuangfan Zhou, Heng Guo, Jinnuo Zhang, Jiang Zhou, Boxin Shi, Zhanyu Ma, Guoying Gu

Abstract

Color photometric stereo enables single-shot surface reconstruction, extending conventional photometric stereo that requires multiple images of a static scene under varying illumination to dynamic scenarios. However, most existing approaches assume ideal distant lighting and Lambertian reflectance, leaving more practical near-light conditions and non-Lambertian surfaces underexplored. To overcome this limitation, we propose a framework that leverages neural implicit representations for depth and BRDF modeling under the assumption of mono-chromaticity (uniform chromaticity and homogeneous material), which alleviates the inherent ill-posedness of color photometric stereo and allows for detailed surface recovery from just one image. Furthermore, we design a compact optical tactile sensor to validate our approach. Experiments on both synthetic and real-world datasets demonstrate that our method achieves accurate and robust surface reconstruction.

Introduction

Photometric Stereo (PS) is a well-established technique for estimating surface normals by observing an object under varying lighting conditions. While traditional PS requires multiple images, Color Photometric Stereo (CPS) utilizes simultaneous Red, Green, and Blue (RGB) illumination to achieve single-shot 3D capture. Despite its potential for dynamic scenes, CPS faces significant challenges in practical environments. Most existing models rely on the "distant light" assumption (parallel rays) and the "Lambertian" assumption (perfectly matte surfaces). In compact setups, such as robotic tactile sensors, lights are positioned near the surface, causing light rays to diverge and intensity to drop rapidly. Furthermore, shiny or metallic (non-Lambertian) surfaces often cause these mathematical models to fail.

Overview of the proposed near-light color photometric stereo framework, showing the setup, input capture, and reconstructed outputs.

Methodology

To address the ill-posed nature of single-shot CPS for non-Lambertian surfaces, this paper introduces a self-supervised framework based on neural implicit representations. The approach is built upon three primary components:

Neural Surface Model

Instead of predicting surface normals directly, the framework employs a Sinusoidal Representation Network (SIREN) coupled with multi-resolution hash encoding to predict a continuous depth map. Surface normals are then derived analytically from this depth map, ensuring physical and geometric consistency across the reconstruction.

Neural BRDF Model

To handle complex reflections, a neural network models the Bidirectional Reflectance Distribution Function (BRDF). By adopting a mono-chromaticity prior—assuming the object consists of a single, uniform material—the network can effectively learn the material's reflectance properties across the entire surface from a single image.

The architecture of the Neural Surface and BRDF Models, illustrating the flow from image coordinates to rendered output for loss calculation.

Crosstalk Correction Module (CCM)

Real-world camera sensors often suffer from color bleeding between channels. A small Multi-Layer Perceptron (MLP) is utilized to pre-process raw images, restoring photometric consistency before they are fed into the reconstruction pipeline.

Experimental Results

The framework was evaluated using both synthetic datasets generated in Blender and real-world data from a custom-built "NearLightTactile" sensor. The mono-chromaticity assumption proved particularly effective for tactile sensing, where robotic "skins" are typically made of uniform materials.

Key findings include:

  • High Accuracy: On synthetic data, the method achieved a Mean Angular Error (MAE) as low as 2.43°, significantly outperforming existing state-of-the-art techniques like LSPS and RGBPS, which often exhibited errors exceeding 11°.
  • Robustness to Specularity: The model maintained high fidelity on challenging materials such as brown metal and white ceramic, successfully handling specular highlights that typically degrade 3D scanning performance.
  • Real-World Application: The custom tactile sensor successfully reconstructed sub-millimeter textures from ancient coins and fabrics using only a single RGB snapshot.

Conclusion and Future Work

This work demonstrates that combining neural implicit representations with a mono-chromaticity prior can turn the ill-posed problem of near-light CPS into a solvable one. While currently limited to single-material objects and requiring calibrated lighting, future research will explore uncalibrated lighting scenarios and multi-material surface reconstruction. The proposed method offers a significant step forward for high-resolution robotic touch and single-shot 3D reconstruction in compact environments.

Summary updated

Abstract

Color photometric stereo enables single-shot surface reconstruction, extending conventional photometric stereo that requires multiple images of a static scene under varying illumination to dynamic scenarios. However, most existing approaches assume ideal distant lighting and Lambertian reflectance, leaving more practical near-light conditions and non-Lambertian surfaces underexplored. To overcome this limitation, we propose a framework that leverages neural implicit representations for depth and BRDF modeling under the assumption of mono-chromaticity (uniform chromaticity and homogeneous material), which alleviates the inherent ill-posedness of color photometric stereo and allows for detailed surface recovery from just one image. Furthermore, we design a compact optical tactile sensor to validate our approach. Experiments on both synthetic and real-world datasets demonstrate that our method achieves accurate and robust surface reconstruction.
BibTeX
@misc{li2026near,
  author = {Zonglin Li and Jieji Ren and Shuangfan Zhou and Heng Guo and Jinnuo Zhang and Jiang Zhou and Boxin Shi and Zhanyu Ma and Guoying Gu},
  title = {{Near-Light Color Photometric Stereo for Mono-Chromatic Non-Lambertian Surfaces}},
  year = {2026},
  eprint = {2601.12666},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.12666},
}
Figure