POSSR

Patch-Based Optimization for Noise-Robust Reconstruction of Specular Surfaces

This repository provides the implementation and data for the paper:

Patch-Based Optimization for Noise-Robust Reconstruction of Specular Surfaces
Saed Moradi, M. Hadi Sepanj, Amir Nazemi, Claire Preston, Anthony M. D. Lee, and Paul Fieguth
IEEE Access, vol. 13, pp. 175884-175894, 2025 📄 PDF | 📊Sample Data

---

🔍 Overview

Reconstructing specular (mirror-like) surfaces from a single camera view is a highly challenging problem in computer vision. This work proposes a patch-based optimization framework that leverages geometric and optical constraints to produce a dense and robust depth map, even under significant noise in point correspondences.

---

🔧 Methodology

We formulate the inverse problem of specular surface reconstruction as a local optimization problem that aligns:

• Normals estimated from reflection geometry
• Normals obtained via local plane fitting

The reconstruction proceeds patch-wise to maintain computational feasibility and robustness.

Problem Description

Figure 1: Surface geometry and reflection overview.

📊 Results

We conducted extensive experiments using synthetically generated data. The proposed method is:

• Robust to noise in both 2D and 3D reflection point correspondences
• Effective with a single camera and a single pattern plane
• Competitive with existing multi-plane methods under ideal conditions

Qualitative Comparison

Figure 2: Qualitative comparison of surface reconstruction results for noise-free (ideal) reflection correspondences.

Quantitative Evaluation

Figure 3: Sensitivity of the reconstructed surface to the noise added to the points on the image plane.

Figure 4: Sensitivity of the reconstructed surface to the noise added to the points on the pattern plane.

Figure 5: Surface reconstruction results of our method in extremely noisy scenarios.

Figure 6: Sensitivity of the reconstructed surface to an erroneous starting point.

---

📂 Repository Structure

├── dataGeneration/                 # A pipeline to generate reflection point correspondences
├── reconstruction/                 # Main MATLAB source code for reconstruction
├── README.md                       # This file

---

📌 Citation

If you find this repository useful in your research, please cite:

@article{moradi2025poss,
  title    =  {Patch-Based Optimization for Noise-Robust Reconstruction of Specular Surfaces},
  author   =  {Moradi, Saed and Hadi Sepanj, M. and Nazemi, Amir and Preston, Claire and Lee, Anthony M. D. and Fieguth, Paul},
  journal  =  {IEEE Access},
  volume   =  {13},
  number   =  {},
  pages    =  {175884-175894},
  year     =  {2025}
}

---

📬 Contact

For questions, please contact Saed Moradi.