perceptual

This is a small library for computing perceptual image hashes. Those hashes allow quickly finding duplicating images.

Currently, there's 5 different hashes implemented here:

• DHash. Fast "difference hash", stored as a uint64 value, meant to be compared using Hamming distance. Uses the idea from here: https://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html
• HHash. Fast, but imprecise "histogram hash". Very reliable against rotations (especially 90/180/270 degrees) and mirrorings, but can easily given false positives or even false negatives (if colors were changed too much).
• PHash. DCT hash, used in phash.org library and described in http://phash.org/docs/pubs/thesis_zauner.pdf
• MHash. Marr hash, used in phash.org library and described in http://phash.org/docs/pubs/thesis_zauner.pdf. Current implementation is very slow and not very reliable; not recommended to use in the current state.
• RHash. Radial hash, used in phash.org library and described in http://phash.org/docs/pubs/thesis_zauner.pdf. Very slow (relative to other hashes) when comparing hashes.

In the future SIFT/SURF feature extraction can be added to the library. This should allow the most precise image matching. However it will also be the slowest one.

API Reference

Installation

nimble install perceptual

Example

import perceptual

let hash1 = dhash("image1.jpg")
let hash2 = dhash("image2.jpg")

let difference = diff(hash1, hash2)

echo "Image Difference: ", difference
if difference < 15:
  echo "Probably the same image!"
elif difference < 25:
  echo "Can be similar"
else:
  echo "Different"