Struct SubPixelRefiner 

pub struct SubPixelRefiner {
    pub half_window: usize,
    pub max_iterations: usize,
    pub epsilon: f64,
}

Sub-pixel corner refinement using parabolic fitting.

After FAST detects corners at integer-pixel positions, this refiner fits a 2-D parabola to the corner response in a small neighbourhood and finds the sub-pixel maximum. This improves localisation accuracy from ~0.5 px to ~0.05 px, which is critical for high-quality homography estimation and camera calibration.

Fields

half_window: usize

Half-size of the refinement window.

max_iterations: usize

Maximum number of Newton iterations.

epsilon: f64

Convergence threshold (pixels).

Implementations

impl SubPixelRefiner

pub fn new(half_window: usize, max_iterations: usize, epsilon: f64) -> Self

Create a new sub-pixel refiner.

pub fn refine( &self, image: &[u8], width: usize, height: usize, keypoints: &[Keypoint], ) -> AlignResult<Vec<Keypoint>>

Refine a set of keypoint positions to sub-pixel accuracy.

Uses a quadratic surface fit: within the window around each keypoint, fit I(x,y) = a*x^2 + b*y^2 + c*x*y + d*x + e*y + f and find the extremum at (-dI/dx, -dI/dy) / Hessian. This works for both corners and peaks/blobs, unlike the structure tensor method which only converges for true corners.

Keypoints too close to the image border are left unmodified.

Errors

Returns an error if the image dimensions are invalid.

Trait Implementations

impl Default for SubPixelRefiner

fn default() -> Self

Returns the “default value” for a type.