Research map / 3D vision / Depth and stereo
Learning to Remove Wrinkled Transparent Film with Polarized Prior
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.
Introduces Film Removal (FR) for wrinkled transparent films over industrial items, with a polarized-camera paired dataset and an end-to-end angle-estimation plus reconstruction network guided by a polarization-based highlight prior.
Problem
Wrinkled transparent films covering industrial products hide information (QR codes, text) and cause failures in downstream recognition; existing highlight removal methods do not model the film's full degradation.
Approach
Physically model film imaging as specular highlight plus diffuse reflection with extra film degradations, estimate per-pixel polarization angle that minimizes highlight via a learned angle network, and use the resulting minimized-highlight prior with four polarized images to reconstruct the original diffuse image.
Key results
- Proposed method achieves mean 10-fold PSNR 36.48 and SSIM 0.9824, versus Restormer 34.32 / 0.9731 and Uformer 31.68 / 0.9426.
- QR code read rate reaches 72.41% for the proposed method versus 27.58% for Restormer and 8.05% for SHIQ (ground truth 100%).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the proposed film-removal framework instead of Restormer | Film Removal · Film Removal, 10-fold cross-validation | PSNR: 34.32 -> 36.48 | Helped |
| Use the proposed film-removal framework instead of Restormer | Film Removal · Film Removal, 10-fold cross-validation | SSIM: 0.9731 -> 0.9824 | Helped |
| Use the proposed film-removal framework instead of Restormer | QR code reading · QR code reading | read rate: 27.58% -> 72.41% | Helped |
| Add polarized information and the Polarization-based Location Model (ablation) | Film Removal framework · Film Removal | PSNR | Helped |
| Add AoP and DoP to the angle estimation network (ablation) | Film Removal framework · Film Removal | PSNR | Helped |
| Add the polarized prior (ablation) | Film Removal framework · Film Removal | PSNR | Helped |
Limitations
- No existing baseline for the new film removal problem; comparisons rely on general reconstruction and highlight removal methods.
- Relies on a specialized four-angle polarization camera, limiting applicability to standard cameras.
Experimental setup
Datasets: Custom polarized film/ground-truth paired dataset
Benchmarks: Custom Film Removal dataset (10-fold cross-validation) QR code reading rate Text OCR
Metrics: PSNR SSIM QR code read rate
Baselines: Uformer Restormer Polar-HR SHIQ
Builds on: Residual Dense Network Malus's Law Stokes parameters
Closest papers by meaning
- Glossy Object Reconstruction with Cost-effective Polarized Acquisition (CVPR 2025 · similarity 0.86)
- PolarFree: Polarization-based Reflection-Free Imaging (CVPR 2025 · similarity 0.85)
- Polarized Color Screen Matting (CVPR 2025 · similarity 0.84)
- PolarMatte: Fully Computational Ground-Truth-Quality Alpha Matte Extraction for Images and Video using Polarized Screen Matting (CVPR 2024 · similarity 0.84)
- Polarization Guided Mask-Free Shadow Removal (AAAI 2025 · similarity 0.83)
- Factored-NeuS: Reconstructing Surfaces, Illumination, and Materials of Possibly Glossy Objects (CVPR 2025 · similarity 0.83)
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.