Yet another movie colors extractor

Overview

This project aims to extract the main colors of each frame of a movie based on a Kmeans clustering of color. Alan Zucconi does a great job explaining the process.

Let us notice that the lab colorspace might give better results as it is supposed to be more perceptually uniform.

Dependencies

• Linux (not tested on macOS, nor Windows)
• Python-3
• OpenCV

How to run it

The required computation is quick heavy. It took around 10 solid minutes to generate a polar plot for a movie of 1h30 on an intel 7700k.

Good old local install

I will supposed that OpenCV is properly installed on your machine. I admit it is a painful process ! I will not redirect you to any tutorials as those tend to be outdated quickly, so good luck. Once OpenCV installed the following set of command will do the trick:

	# Debian/Ubuntu
	$ apt install python-3 python-pip
	$ pip install --upgrade virtualenv
	$ git clone https://github.com/adrz/movie-main-colors
	$ cd movie-main-colors
	$ virtualenv -p python3 env
	$ . env/bin/activate
	$ pip install -r requirements.txt
	$ python main-colors.py -i "/home/user/movie.mp4" -o "/home/user/output.png"

Docker

In case you are not a linux wizard, I recommend you to go with docker. First install docker: see instructions, then to generate a polar plot of the movie 'movie.mp4':

	# Debian/Ubuntu
	$ python main-colors-docker.py -i "/home/user/movie.mp4" -o output.png

The result will be output in the folder: 'movie-main-colors/results/'.

Parameters/Options

  -h, --help            show this help message and exit
  -i, --input_file      input movie file (movie.mp4)
  -o, --ouput_file      output image file (colors-movie.png)
  -a, --alg             algorithm used to cluster colors (choice: sklearn, cv,
                        gaussian), default: cv
  -c, --colorspace      colorspace in which the clusters are computed (choice: hsv, hls,
  luv, lab), default: lab
  -n, --n_colors        number of colors to extract, default: 3
  --normalize           normalize the colors before clustering (choice: 0, 1), default: 1
  -x, blur_xy           width of the smoothing kernel, default: 5
  -s, --saturate        color saturation factor, a value above 1 make color more ''vivid'', default: 1
  -t, --type            type of plot (choice: polar, bar), default: polar

Some other movie colors extractors/info

Before jumping into other project, I highly recommend you a very interesting analysis of uses of color in movies. Some guys on twitter also share homemade color palettes based on a single iconic frame of the movies.

Several movie colors extractors exist in the wild. Most of them produces a horizontal or vertical ''barcode'' produced by a concatenation of the average color of individual frames. It is often less computationally intensive and rely on ffmpeg: movie barcodes generator. You could even buy a poster of your favorite movie thecolorinmotion.com. The most impressive work is by far a bachelor graduation project called cinemetrics.