Principle:Kornia Kornia Inlier Extraction
| Knowledge Sources | |
|---|---|
| Domains | Vision, Geometry |
| Last Updated | 2026-02-09 15:00 GMT |
Overview
Technique of filtering verified correspondences from raw feature matches using a geometric inlier mask.
Description
After RANSAC geometric verification, an inlier mask identifies which feature correspondences are consistent with the estimated geometric model. Inlier extraction applies this boolean mask to the original keypoint pairs to obtain only the verified correspondences. These clean correspondences serve as reliable input for downstream tasks like 3D reconstruction, camera pose estimation, and image stitching. The extraction is a simple but critical step that bridges robust estimation and geometric computation.
Usage
Use immediately after RANSAC to obtain the set of geometrically consistent correspondences. Required before any task that assumes clean point correspondences.
Theoretical Basis
Given keypoint pairs (kp1, kp2) and inlier mask m from RANSAC:
kp1_verified = kp1[m]
kp2_verified = kp2[m]
The inlier ratio |m_true| / |m| indicates matching quality.