Feature Request: Curve-Constrained Vertical Warping for Geometric Correction
Context
In the current KH-9 image processing pipeline, mosaicking introduces significant geometric distortions (notably vertical bowing and shear artifacts). To correct these distortions, we currently rely on a combination of RANSAC-based boundary detection (top and bottom edges) and a TPS-based or blockwise remapping strategy.
However, this approach can be overly flexible and computationally heavy, and may introduce unnecessary deformation artifacts outside the region of interest.
Proposed Feature
Implement a Curve-Constrained Vertical Warping (CCVW) transformation model.
This method uses two boundary curves (top and bottom) to define a normalized vertical coordinate system, and performs a deterministic, monotonic remapping between source and target geometry.
Advantages Over Current Approach
Compared to TPS:
- Reduced computational cost
- No global system solving
- More stable in cases with quasi-1D deformation structure
- Better interpretability
Compared to affine models:
- Spatially varying deformation along x-axis
- Captures non-linear bowing effects typical of KH-9 mosaicking
Exemple of this transformation on regular grid
Note : this case is just to illustrate the effect transformation so the 2 lines are really curved but usually tranformation will be softer.
Curve-constrained Vertical Warping
Thin plate spline

Feature Request: Curve-Constrained Vertical Warping for Geometric Correction
Context
In the current KH-9 image processing pipeline, mosaicking introduces significant geometric distortions (notably vertical bowing and shear artifacts). To correct these distortions, we currently rely on a combination of RANSAC-based boundary detection (top and bottom edges) and a TPS-based or blockwise remapping strategy.
However, this approach can be overly flexible and computationally heavy, and may introduce unnecessary deformation artifacts outside the region of interest.
Proposed Feature
Implement a Curve-Constrained Vertical Warping (CCVW) transformation model.
This method uses two boundary curves (top and bottom) to define a normalized vertical coordinate system, and performs a deterministic, monotonic remapping between source and target geometry.
Advantages Over Current Approach
Compared to TPS:
Compared to affine models:
Exemple of this transformation on regular grid
Note : this case is just to illustrate the effect transformation so the 2 lines are really curved but usually tranformation will be softer.
Curve-constrained Vertical Warping
Thin plate spline