Detect repeated patterns

Locate repeated structures with Template Match, then turn score peaks into a coordinate table with Find Peaks. This detects locations; it does not segment object boundaries or create labels.

New in 0.16.0a3

These CPU-only nodes and two examples require 0.16.0a3 or newer. Matching uses one fixed template size and orientation.

Start with a known-answer example

Choose Open example… from the workflow actions menu, then open Detection & Peaks:

Example	Expected result at the saved settings
Repeated 2D Pattern Detection	Five asymmetric patterns, including two centers 12 pixels apart.
Anisotropic 3D Pattern Detection	Four volumetric patterns; a nearby pair is 6.4 micrometres apart.

Use Calculate all, then inspect Find Peaks. Both examples explicitly select timepoint 1 and the Repeated pattern channel (1), crop one complete template, connect the valid-support mask, and plot retained scores. The other channel is independent noise; the other timepoint has a different arrangement. Canvas notes give the known center coordinates. A clipped border copy and a deliberately absent site should not appear in the detection table.

Build the detection branch

1. Check the source axes, pixel/voxel spacing and units. Use Select Axis Slice for the intended timepoint and Extract Channel for the intended channel. The resulting image must have explicit YX or ZYX axes.
2. Branch that selected image into Crop Stack. Crop tightly around one complete, distinctive example of the desired pattern. Keep Z for a 3D template; a projection is a different scientific input.
3. Connect the selected full image to Template Match → Search image and the cropped image to Template image. The template may have a different origin but must have matching sampling, spatial rank and axis order.
4. Calculate and inspect Match scores and Valid scores. The score map is smaller than the source and is calibrated at template centers, not at the top-left template corner.
5. Under Template Match's Next step, choose Add Find Peaks. This connects both outputs and enables Use valid mask, without calculating. Show Find Peaks selects an existing continuation. Set Minimum value and Minimum separation deliberately. Choose physical distance when unequal voxel spacing matters.
6. Calculate and inspect the detection table. Under Detection review, choose Inspect detections on source… for the read-only coordinate overlay. Review detections throughout Z, including faint, nearby and border structures. Connect the table to Plot Results to explore score, or export the table through the existing results controls.

Click a table row or marker to identify the same detection. In 3D, use Z plane (zero-based); the overlay shows one plane, not a projection. Refresh view loads an already calculated result and does not calculate the workflow. If the source or result changes, recalculate before interpreting the overlay.

Raise the score cutoff to remove weaker matches; lower it only while reviewing false detections. Raising minimum separation can suppress real nearby objects. The 2D example changes from five to four detections at 13 pixels; the 3D example changes from four to three at 7 micrometres.

A high correlation is not a probability

The template's own location is included when it comes from the search image. Its near-one score is not independent validation. Patterns with changed scale, orientation, shape or contrast may be missed, while unrelated structures may match. Validate choices against independent representative annotations, not only the bundled synthetic examples.

See Template matching and peaks for exact distance, border, tie, numeric and table-coordinate rules.