Perceptual hashing for images. Implemented as a C++ program connected to Elixir via a port.

Requires Magick++, the ImageMagick C++ API. Check your package manager or the ImageMagick downloads page.

For use in a mix project, add ExPerHash to the mix.exs dependencies:

def deps do
  [{:experhash, github: "kemonomachi/experhash"}]
end

Download by using:

$ mix deps.get

Run mix compile to build both the Elixir and the C++ code.

iex> {:ok, srv} = ExPerHash.start_link
{:ok, #PID<0.91.0>}

iex> {:ok, hash1} = ExPerHash.dd_hash_128 srv, "some/image.png"
{:ok, <<140, 173, 167, 143, 157, 141, 14, 39, 77, 9, 3, 135, 23, 49, 25, 89>>}

iex> {:ok, hash2} = ExPerHash.dd_hash_128 srv, "some/other/image.png"
{:ok, <<140, 173, 167, 143, 157, 141, 14, 39, 77, 9, 0, 135, 23, 49, 25, 89>>}

iex> ExPerHash.hamming_distance hash1, hash2
{:ok, 2}

Three hash functions are available: aHash, dHash and ddHash.

aHash by dr. Neal Krawitz creates a hash based on the average of the low frequencies of an image.

dHash by dr. Neal Krawitz and David Oftedal tracks gradients instead. Hash bits are set in order from left to right (most significant to least significant).

ddHash is a double dHash, one row-wise and one column-wise.

All hash functions return {:ok, hash} on success, where hash is a binary with four possible sizes. For aHash and dHash, the sizes are 64, 256, 1024, and 4096 bits. For ddHash, the sizes are 128, 512, 2048, and 8192 bits. Each function name ends with the hash size, like a_hash_64 and dd_hash_128.

On error, all functions return {:error, {error_type, reason}.

The hamming_distance/2 function compares two hashes and return the number of bits that differ. Low distance means similar images.

Copyright © 2015-2018 Ookami Kenrou <[email protected]>

This work is free. You can redistribute it and/or modify it under the terms of the Do What The Fuck You Want To Public License, Version 2, as published by Sam Hocevar. See the LICENSE file or the WTFPL homepage for more details.