matching op• Data kinds: image → match
• Call: fullseye.apply(img, "shape_locate", a=0.5, b=0.5) (the 2-D model is one image plus two scalar knobs a,b∈[0,1])
• HALCON equivalent: find_shape_model (the HALCON reference is a useful guide to its meaning and parameters)
• GPU: none (CPU reference implementation only)
*The figure is the actual output on a synthetic 128×128 input. Left: input, right: output. Point clouds are drawn as a top-down scatter (brightness = z), 1-D series as a line plot, volumes as the maximum-intensity projection along z, videos as the middle frame, complex images as magnitude; return values that are not pictures are shown as the values themselves.*
*Knob a does not change the output (measured: identical at 0.1 / 0.5 / 0.9).*
*Knob b does not change the output (measured: identical at 0.1 / 0.5 / 0.9).*
Template matching that accounts for rotation (shape-based matching). Equivalent to HALCON's find_shape_model (Find the best matches of a shape model in an image.).
a, b are unused — as with _ncc_locate, the template is pre-registered with set_match_template. The template is rotated from 0°〜330° in 30° steps, _ncc_map (NCC) is computed for each, and the best correlation over all positions and all angles is returned as [correlation, y, x, angle]. Because the angle step is coarse (30°), it does not reach continuous angular precision like HALCON's find_shape_model — it is for rough orientation detection (to choose the step and refine below it, use matching.shape_locate_fine). If no template is set, [0,0,0,0].
The corners of the rotated template are left out of the correlation (★2026-11-09 op audit). Rotating with reshape=False brings the corners in from outside, so those pixels are masked out and Pearson correlation is computed over the remaining pixels only (matching._MaskedNCC). Previously the zero-filled corners were correlated as part of the template, so on images whose background is not 0 the score dropped structurally even at the correct angle (background 0.5, 30° rotation: 0.7794 → 0.981 after the fix). Only positions where every valid pixel of the rotated template lies inside the image are searched (at angle 0 this is the same full-overlap rule as ncc_locate, and the score agrees to rounding). An image containing NaN / inf raises FsValueError (on the pipeline path ops.RT / run_stages the 4-element [0,0,0,0] is returned and recorded in fs.fallbacks()).
• gallery2d_contour_measure family guide
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
The program below has been verified to run (same input as the figure). In Studio's help this block becomes buttons that load and run it on the spot.
shape_locate 0.50 0.50
▸ Load this pipeline · Load & run
• gallery2d_contour_measure — py -3.11 examples/gallery2d_contour_measure.py
• poc_template_tracking — py -3.11 examples/poc_template_tracking.py
match as input)matching)*Provenance: ops.py — 2D operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.