Local anomaly inspection learned from normal machining textures
A new machining mark may match neither known scratch nor burr classes. Inspection relying only on labelled defect classes can miss an unknown departure from normal texture.
Store the anomaly score, above-threshold area and raw ROI together. Localise the hot region for operator review; any new defect class is defined later if needed.
- 01 Normal reference
- 02 New image
- 03 Local comparison
- 04 Score threshold
- 05 Send for review
Select the illustrative reference gallery of normal examples only.
- Local anomaly score
- —
- Above-threshold area
- —
- Illustrative score threshold
- —
Scenes are illustrative simulations; they are not customer references or performance guarantees.
Solution approach
Local features of normal examples are compared with the new ROI. Raw evidence, normal reference and a continuous anomaly map appear separately; the score threshold is not a defect class. Register machining direction and display the normal texture gallery beside the new ROI difference. A local hot region goes for review without a defect name.
Process steps
- 01
Normal reference
Select the illustrative reference gallery of normal examples only.
- 02
New image
Image the new sample under the same lighting and registration conditions.
- 03
Local comparison
Show local feature deviation as a continuous heatmap.
- 04
Score threshold
Show the local score and above-threshold area as separate criteria.
- 05
Send for review
Hold an anomalous example for the operator without inventing a defect type.
Related applications
Let us assess your application together
We test your sample in our laboratory and report on resolution, optics and lighting selection.