Struct TemplateMatcher 

pub struct TemplateMatcher { /* private fields */ }

Grayscale normalized cross-correlation template matcher.

evaluate slides the template over the region and returns the best match as DetectorOutput::Score in 0.0..=1.0 (negative correlation clamps to 0). Naive O(region * template) search — fine for the small regions and icons this library targets; optimize (integral images / FFT) only if profiling ever demands it.

Degenerate (flat) inputs fall back to mean comparison: a flat template scores 1.0 on a window of the same mean brightness, else 0.0.

NCC is contrast-invariant: a dim copy of the pattern scores as high as a bright one, so the score says nothing about absolute brightness/contrast. A template larger than the watched region scores 0.0 (“not found”) rather than erroring — check your region size if a template watcher never fires.

Implementations

impl TemplateMatcher

pub fn from_luma( width: u32, height: u32, luma: Vec<u8>, ) -> Result<TemplateMatcher, DetectorError>

Build from a row-major grayscale buffer. Errors if luma.len() does not equal width * height or either dimension is zero.

pub fn from_png_bytes(bytes: &[u8]) -> Result<TemplateMatcher, DetectorError>

Build from encoded image bytes (PNG). Color images are converted to grayscale; matching is luma-only. Alpha is ignored (not composited); export icons with an opaque background.

Trait Implementations

impl Detector for TemplateMatcher

fn evaluate( &mut self, view: &FrameView<'_>, ) -> Result<DetectorOutput, DetectorError>

fn is_heavy(&self) -> bool

Heavy detectors (OCR, ML inference) are routed to the blocking thread pool by the watcher loop so they never stall the async scheduler. Cheap pixel-math detectors keep the default false.