Match
The match endpoint is the core of Akator Image Matcher. Check a user upload against all images in a collection and get matches ranked by similarity.
Endpoint reference: the exact path, parameters, request body and response schema are generated from the live OpenAPI spec — see the interactive API reference. This page focuses on how to interpret the results.
Understanding the Metrics
Each match comes with several similarity metrics so you can make informed decisions:
| Metric | Best For | Interpretation |
|---|---|---|
ssim_score | Overall similarity | ≥0.9 = very similar, ≥0.8 = similar, <0.7 = different |
match_percent | Feature matching | Higher = more matching SIFT keypoints found |
iou_percent | Crop detection | Lower values may indicate cropping |
wasserstein_distance | Color changes | Higher values indicate color modifications |
Recommended Thresholds
| SSIM Score | Interpretation | Recommended Action |
|---|---|---|
| ≥ 0.95 | Near-identical or identical | Block automatically |
| 0.85 - 0.95 | Very similar (minor edits) | Block or flag for review |
| 0.70 - 0.85 | Similar (significant edits) | Flag for manual review |
| < 0.70 | Potentially unrelated | Review with caution |
Integration Best Practices
- Start with a moderate threshold: Use
min_ssim_score=0.7initially and adjust based on your false positive/negative rate. - Use multiple metrics: Combine SSIM with IoU to detect cropped versions more reliably.
- Handle matches gracefully: Provide users with a clear explanation and appeals process, as required by UrhDaG.
- Log all checks: Keep records of all match checks for compliance documentation.
- Consider async processing: For high-volume uploads, check images asynchronously to avoid slowing down the upload flow.