/dhash

Python library to calculate the difference hash (perceptual hash) for a given image, useful for detecting duplicates

Primary LanguagePythonMIT LicenseMIT

dhash

dhash on PyPI (Python Package Index) GitHub Actions Tests

dhash is a Python library that generates a "difference hash" for a given image -- a perceptual hash based on Neal Krawetz's dHash algorithm in this "Hacker Factor" blog entry.

The library is on the Python Package Index (PyPI) and works on both Python 3 and Python 2.7. To install it, fire up a command prompt, activate your virtual environment if you're using one, and type:

pip install dhash

The algorithm to create a difference hash is very simple:

  • Convert the image to grayscale
  • Downsize it to a 9x9 thumbnail (size=8 means an 8+1 by 8+1 image)
  • Produce a 64-bit "row hash": a 1 bit means the pixel intensity is increasing in the x direction, 0 means it's decreasing
  • Do the same to produce a 64-bit "column hash" in the y direction
  • Combine the two values to produce the final 128-bit hash value

The library defaults to producing a size 8 dhash, but you can override this easily by passing size=N as a keyword argument to most functions. For example, you can produce a more accurate (but slower to work with) dhash of 512 bits by specifying size=16.

I've found that dhash is great for detecting near duplicates (we found dupes using a size 8 dhash with a maximum delta of 2 bits). But because of the simplicity of the algorithm, it's not great at finding similar images or duplicate-but-cropped images -- you'd need a more sophisticated image fingerprint if you want that. However, the dhash is good for finding exact duplicates and near duplicates, for example, the same image with slightly altered lighting, a few pixels of cropping, or very light photoshopping.

To use the dhash library, you need either the wand ImageMagick binding or the Pillow (PIL) library installed. Pick one and stick with it -- they will produce slightly different dhash values due to differences in their grayscale conversion and resizing algorithms.

If you have both libraries installed, dhash will use wand by default. To override this and force use of Pillow/PIL, call dhash.force_pil() before using the library.

To produce a dhash value using wand:

import dhash
import wand.image

with wand.image.Image(filename='dhash-test.jpg') as image:
    row, col = dhash.dhash_row_col(image)
print(dhash.format_hex(row, col))

To produce a dhash value using Pillow:

import dhash
from PIL import Image

image = Image.open('dhash-test.jpg')
row, col = dhash.dhash_row_col(image)
print(dhash.format_hex(row, col))

If you have your own library to convert an image to grayscale and downsize it to 9x9 (or 17x17 for size=16), you can pass dhash_row_col() a list of integer pixel intensities (for example, from 0 to 255). For example:

>>> import dhash
>>> row, col = dhash.dhash_row_col([0,0,1,1,1, 0,1,1,3,4, 0,1,6,6,7, 7,7,7,7,9, 8,7,7,8,9], size=4)
>>> format(row, '016b')
'0100101111010001'
>>> format(col, '016b')
'0101001111111001'

To produce the hash value as a 128-bit integer directly, use dhash_int(image, size=N). To format the hash value in various ways, use the format_* functions:

>>> row, col = (13962536140006260880, 9510476289765573406)
>>> dhash.format_bytes(row, col)
b'\xc1\xc4\xe4\xa4\x84\xa0\x80\x90\x83\xfb\xff\xcc\x00@\x83\x1e'
>>> dhash.format_hex(row, col)
'c1c4e4a484a0809083fbffcc0040831e'

To compute the number of bits different (hamming distance) between two hashes, you can use the get_num_bits_different(hash1, hash2) helper function:

>>> import dhash
>>> dhash.get_num_bits_different(0x4bd1, 0x5bd2)
3

You can also use dhash to generate the difference hash for a specific image from the command line:

$ python -m dhash dhash-test.jpg
c1c4e4a484a0809083fbffcc0040831e

$ python -m dhash --format=decimal dhash-test.jpg
13962536140006260880 9510476289765573406

# show the 8x8 row and column grids
$ python -m dhash --format=matrix dhash-test.jpg
* * . . . . . *
* * . . . * . .
* * * . . * . .
* . * . . * . .
* . . . . * . .
* . * . . . . .
* . . . . . . .
* . . * . . . .

* . . . . . * *
* * * * * . * *
* * * * * * * *
* * . . * * . .
. . . . . . . .
. * . . . . . .
* . . . . . * *
. . . * * * * .

# compute the bit delta between two images
$ python -m dhash dhash-test.jpg similar.jpg
1 bit differs out of 128 (0.8%)

Read the code in dhash.py for more details – it's pretty small!

dhash was written by Ben Hoyt and is licensed with a permissive MIT license (see LICENSE.txt).