chess-corners-core

Core primitives for computing ChESS responses and extracting subpixel chessboard corners.

This crate implements:

• 16-sample ChESS rings (ring module) at radii 5 and 10.
• Dense response computation on 8-bit grayscale images (response module).
• Thresholding, non-maximum suppression, and pluggable subpixel refinement (detect + refine modules).
• Conversion from raw peaks to rich corner descriptors (descriptor module).

Feature flags:

• std (default) – use the Rust standard library; disabling this yields no_std + alloc.
• rayon – parallelize response computation over image rows.
• simd – enable portable-SIMD acceleration of the response kernel (nightly only).
• tracing – emit structured spans around response and detector code for profiling.

Basic usage:

use chess_corners_core::{detect::find_corners_u8, ChessParams, RefinerKind};

fn detect(img: &[u8], w: usize, h: usize) {
    let mut params = ChessParams::default();
    // Default = center-of-mass refinement on the response map.
    let corners = find_corners_u8(img, w, h, &params);
    println!("found {} corners", corners.len());

    // Opt into Förstner or saddle-point refiners on the image intensities:
    params.refiner = RefinerKind::Forstner(Default::default());
    let refined = find_corners_u8(img, w, h, &params);
    println!("found {} corners with Förstner", refined.len());
}

For a higher-level, image-friendly API (including multiscale detection and an optional CLI), see the chess-corners crate in this workspace.