Research map / 3D vision / Depth and stereo
Generalized Closed-Form Formulae for Feature-Based Subpixel Alignment in Patch-Based Matching
Laurent Valentin Jospin, Hamid Laga, Farid Boussaïd, Mohammed Bennamoun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace parabola cost interpolation with barycentric feature interpolation | ZNCC subpixel refinement · Middlebury 2014 stereo (1D, 5x5 window) | MAE (px): 0.15 -> 0.12 | Improved |
| replace parabola cost interpolation with equiangular or Shimizu-Okutomi to reduce pixel locking | ZNCC 5x5 cost-based refinement · Middlebury 2014 stereo (1D) | pixel-locking SNR (dB): -13.14 -> -26.68 | Improved |
| use predictive interpolation (left and right features together) instead of barycentric | ZNCC feature-based refinement · Middlebury 2014 stereo (1D, 5x5 window) | MAE (px): 0.12 -> 0.15 | Hurt |
| enlarge correlation window from 5x5 to 11x11 (ablation) | ZNCC barycentric feature refinement · Middlebury 2014 stereo (1D) | MAE (px): 0.12 -> 0.15 | Hurt |
| use feature-based instead of cost-based refinement for SAD | SAD cost · Middlebury 2014 stereo (1D, 5x5 window) | MAE (px): 0.18 -> 0.19 | Hurt |
| use barycentric feature refinement instead of parabola for MedS | MedS cost · Middlebury 2014 stereo (1D, 5x5 window) | MAE (px): 0.20 -> 0.27 | Hurt |
| replace isotropic parabola with Queen split feature refinement | ZNCC subpixel refinement · Middlebury optical flow (2D, 11x11 window) | mean distance (px): 0.22 -> 0.16 | Improved |
| replace isotropic parabola with Queen split feature refinement | ZNCC subpixel refinement · Middlebury optical flow (2D, 5x5 window) | mean distance (px): 0.26 -> 0.19 | Improved |
| replace isotropic with anisotropic parabola cost refinement | ZNCC cost-based refinement · Middlebury optical flow (2D, 11x11 window) | mean distance (px): 0.22 -> 0.27 | Hurt |
| consider all four corners symmetrically (Queen) instead of splitting features | ZNCC feature refinement · Middlebury optical flow (2D, 11x11 window) | mean distance (px): 0.16 -> 0.22 | Hurt |
| use Queen split feature refinement for MedS instead of isotropic parabola | MedS cost · Middlebury optical flow (2D, 11x11 window) | mean distance (px): 0.25 -> 0.34 | Hurt |
| enlarge correlation window from 5x5 to 11x11 (ablation) | ZNCC Queen split feature refinement · Middlebury optical flow (2D) | mean distance (px): 0.19 -> 0.16 | Improved |
Closest papers by meaning
- PatchMatch Filter: Edge-Aware Filtering Meets Randomized Search for Visual Correspondence (TPAMI 2016 · similarity 0.88)
- Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (JMLR 2016 · similarity 0.86)
- Temporally Consistent Stereo Matching (ECCV 2024 · similarity 0.86)
- A Memory- and Accuracy-Aware Gaussian Parameter-Based Stereo Matching Using Confidence Measure (TPAMI 2019 · similarity 0.86)
- IGEV++: Iterative Multi-Range Geometry Encoding Volumes for Stereo Matching (TPAMI 2025 · similarity 0.86)
- Any-Stereo: Arbitrary Scale Disparity Estimation for Iterative Stereo Matching (AAAI 2024 · similarity 0.85)
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.
