Struct WeightedHomographyRefiner 

pub struct WeightedHomographyRefiner {
    pub sigma: f64,
    pub iterations: usize,
}

Weighted least squares homography refiner.

After RANSAC identifies inliers, this refiner computes a more accurate homography by weighting each correspondence inversely by its reprojection error. Points closer to the model contribute more, producing estimates that are more robust to near-outlier noise.

The weighting function is a Cauchy (Lorentzian) kernel:

w(e) = 1 / (1 + (e / sigma)^2)

This is iterated several times (IRLS - Iteratively Reweighted Least Squares) to converge to a robust M-estimate.

Fields

sigma: f64

Scale parameter for the Cauchy kernel.

iterations: usize

Number of IRLS iterations.

Implementations

impl WeightedHomographyRefiner

pub fn new(sigma: f64, iterations: usize) -> Self

Create a new weighted homography refiner.

pub fn refine( &self, initial: &Homography, matches: &[MatchPair], ) -> AlignResult<Homography>

Refine a homography using iteratively reweighted least squares.

initial is the RANSAC-estimated homography. matches is the full set of inlier correspondences.

Errors

Returns an error if there are fewer than 4 matches or the system is degenerate.

Trait Implementations

impl Default for WeightedHomographyRefiner

fn default() -> Self

Returns the “default value” for a type.