Research map / 3D vision / Depth and stereo
Coherence As Texture - Passive Textureless 3D Reconstruction by Self-interference
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes using spatial coherence of natural light as a self-interference texture to passively recover depth of textureless surfaces via a Michelson-style interferometer with a flipped copy and tilted orthographic camera.
Problem
Passive depth estimation (stereo, defocus) fails on textureless objects because there are no intensity features to establish correspondences; prior passive methods recover only normals or are impossible.
Approach
Treat the spatial coherence of natural illumination as a texture: a self-interferometer interferes the scene image with a horizontally flipped copy, so only points on the central column produce interference; the peak position in each row, measured via a tilted orthographic camera, gives a disparity shift that maps linearly to depth.
Key results
- Lab prototype reconstructs textureless planar and stair targets under sunlight and indoor light; stair spacing measured ~0.99 mm vs 0.9 mm ground truth.
- Analysis predicts ~38 resolvable depth planes for indoor light (coherence 4 µm) and ~500 for sunlight (coherence 60 µm).
- Reconstruction error of a metal pen cylinder is ~0.1 mm at 108 µm coherence and increases as coherence length decreases (0.28-0.3 mm at 27-16%).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| reduce illumination coherence length from Δc = 54 µm to Δc = 27 µm (ablation) | Coherence As Texture · Cylindrical pen | average reconstruction error: 0.13 -> 0.28 | Hurt |
| reduce illumination coherence length from Δc = 27 µm to Δc = 16 µm (ablation) | Coherence As Texture · Cylindrical pen | average reconstruction error: 0.28 -> 0.3 | Hurt |
| reduce illumination coherence length from Δc = 108 µm to Δc = 54 µm (ablation) | Coherence As Texture · Cylindrical pen | average reconstruction error: 0.14 -> 0.13 | Helped |
| vary illumination coherence length from Δc = 108 µm to Δc = 16 µm (ablation) | Coherence As Texture · Cylindrical pen | average reconstruction error: 0.14, 0.13, 0.28, 0.3 | Mixed |
Surprising findings
- Depth of a uniform textureless plane can be passively recovered under natural illumination, contrary to the widely accepted view that it is impossible.
- Only a single central-column point per row contributes signal, so the approach effectively scans one column at a time.
Limitations
- Depends critically on the coherence length of the illumination; extended light sources give too-small coherence for reliable depth.
- Subsurface scattering reduces interference contrast; non-metallic objects degrade reconstruction.
- Specular or view-dependent objects degrade performance because the two arms receive unequal intensities.
Experimental setup
Metrics: interference contrast peak position variance depth error (mm) depth resolution depth range
Baselines: stereo matching shape from shading shape from polarization structured light time-of-flight
Builds on: Michelson interferometer phase-shifting interferometry FINCH COACH Kotwal et al. OCT under sunlight
Closest papers by meaning
- Depth Estimation by Combining Binocular Stereo and Monocular Structured-Light (CVPR 2022 · similarity 0.84)
- Multi-View 3D Reconstruction of a Texture-Less Smooth Surface of Unknown Generic Reflectance (CVPR 2021 · similarity 0.84)
- Inferring Light Fields From Shadows (CVPR 2018 · similarity 0.83)
- Beyond Volumetric Albedo -- A Surface Optimization Framework for Non-Line-Of-Sight Imaging (CVPR 2019 · similarity 0.83)
- 3D Gaussian Flats: Hybrid 2D/3D Photometric Scene Reconstruction (NeurIPS 2025 · similarity 0.83)
- DCDepth: Progressive Monocular Depth Estimation in Discrete Cosine Domain (NeurIPS 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.