image-intensities 0.0.10.dev7

Calculate image intensities, a database friendly alternative to image hashing..

image_intensities

Python implementation of the great derpibooru/image_intensities.

The algorithmic overview is to convert the image into the yuv colorspace, drop the u and v components, and then average the y component over 4 evenly-spaced rectangles on the image dimensions.

Usage

from image_intensities import png_intensities, jpg_intensities, image_intensities, Intensities

# Let's calculate some values
luma_a = png_intensities('/absolute/path/to/954482.png')  # image can ge found in the tests folder.
luma_b = jpg_intensities('/absolute/path/to/2544057.jpg')  # image can ge found in the tests folder.
# if you don't know the file type, you can have it picked up with the `mimetype` module:
luma_b = image_intensities('/absolute/path/to/2544057.jpg')  # image can ge found in the tests folder.

# returns something like
luma_a = Intensities(nw=35.832628091300684, ne=10.513891063388744, sw=20.76546499989676, se=20.831389937866714)
luma_b = Intensities(nw=8.284639125603292, ne=8.466390155604937, sw=22.851929679674072, se=23.26008498727572)

# You can compare them, exact match (which is probably not what you want, see .compare(…) below)
luma_a == luma_b
# -> False

# or a distance comparison
luma_a.compare(luma_b, distance=0.25)
# -> False  # those two images are very different, after all

A good distance is usually between 0.2 and 0.5, the default is 0.25.

Pure Python

There's also a pure python version, which thanks to PIL/Pillow supports about every image type:

⚠️ Note: You need to install Pillow (pip install Pillow) for this to work, see Dependencies below.
⚠️ Note: For animated image types it will probably use the first frame.

from image_intensities.pure_python import png_intensities, jpg_intensities, image_intensities, Intensities

luma = image_intensities('/path/to/image.gif')

# returns something like
luma == Intensities(nw=0.42, ne=0.44, sw=0.58, se=0.69)

Dependencies

C-Extension

MacOS

brew install libpng

(tested to work with 1.6.37)

Ubuntu

(There's also a version for Dockerfiles available)

sudo apt-get update
sudo apt-get install libpng-dev

Dockerfile (Ubuntu based)

apt-get update -y && apt-get install -y libpng-dev && apt-get clean && rm -rfv /var/lib/apt/lists/*

⚠️ Make sure you pull a recent release of the python docker image (even if it's an older python version!). Especially if you're getting errors complaining about a png_set_longjmp_fn function when you try to use it. As time of writing the 2-month old version did not work, but the newest releases (python:3.6: 6ac87e65b6d0, pythong:3.9: 1b33974176a3) ones have that fixed.

Pure python algorithm (optional)

This one is usually not needed, unless you wanna benefit from pillow's ability to read the strangest image formats. But probably it's faster to write it to disk as either .jpg or .png and use the native C versions as above.

pip install Pillow

Installation

See dependencies above, and make sure those are installed as needed.

Install via pip

pip install image_intensities

Mac OS:

CPPFLAGS='-I/usr/local/include/' LDFLAGS='-L/usr/local/lib/' pip install image_intensities

From source

E.g. you checked out this repository:

python setup.py install

Mac OS:

CPPFLAGS='-I/usr/local/include/' LDFLAGS='-L/usr/local/lib/' python setup.py install

Minimal installation example

Using docker as it already has a working libpng-dev:

docker pull python:3.9
docker run -it --rm python:3.9 bash
pip install image_intensities

# test import
python -c"import image_intensities as it; print(it._intensities.ffi)"

# test with a png
wget https://derpicdn.net/img/download/2015/8/9/954482.png -O /tmp/954482.png
python -c"from image_intensities import png_intensities; print(png_intensities('/tmp/954482.png'))"

# test with a jpg
wget https://derpicdn.net/img/download/2021/2/4/2544057.jpg -O /tmp/2544057.jpg  # we need to use an absolute path!
python -c"from image_intensities import jpeg_intensities; print(jpeg_intensities('/tmp/2544057.jpg'))"