Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update README with note on dihedral pdq hashes #1694

Merged
merged 1 commit into from
Nov 11, 2024
+6 −0
Merged

Update README with note on dihedral pdq hashes #1694

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions pdq/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,12 @@ Before evaluating the results on your own to choose the thresholds that work for
* **Distance Threshold to consider two hashes to be similar/matching**: <=31
* **Quality Threshold where we recommend discarding hashes**: <=49

## Note on Dihedral PDQ Hashes

The PDQ hashing algorithm is easily capable of producing eight "dihedral" hashes (one for each 90 degree rotation and one for each flip across a horizontal, vertical or diagonal axis). However, PDQ does not guarantee exact rotational invariance. Small variations can occur in the hash values for each rotation due to how PDQ processes the image’s grid alignment in its DCT (Discrete Cosine Transform) phase.

For example, two rotated versions of an image can have a slightly different set of eight dihedral hashes. Selecting a "minimal" hash from these transformations (e.g., lexicographically) may yield inconsistent results because of these minor bit differences. For each image, if we select the minimal hash, there’s no guarantee that the same hash will be selected across different rotations. These inconsistencies arise when small bit variations lead to a different hash being identified as "minimal" for each rotation. For a clearer example, check this issue: ([https://github.com/facebook/ThreatExchange/issues/1676#issuecomment-2466331532](https://github.com/facebook/ThreatExchange/issues/1676#issuecomment-2466331532)).

## Contact

[email protected]