/colorcorrectionmatrix

Compute color correction matrices and develop raw images in Python.

Primary LanguagePythonMIT LicenseMIT

Introduction

This project aims to quickly and easily generate and apply a 3x3 color correction matrix (CCM).

Disclaimer

This is a heavily-modifed fork of lighttransport/colorcorrectionmatrix. I (dirtbirb) have added a color extraction tool (extractColor.py), extensively modified two of the original tools (computeCCM.py and correctColor.py), and removed much of the original content. I want to give credit where credit is due, but please don't blame lighttrasnport if this does something horrible to your computer. Or me, for that matter (see MIT license).

Contents

This repository contains three related tools:

  • extractColor: Find and extract color values from a raw or processed image containing a 24-chip ColorChecker grid.
  • computeCCM: Compute the color correction matrix (CCM) to convert from one set of colors to another other in XYZ space.
  • correctColor: Apply a CCM and other corrections to a raw or processed image.

Each tool is described in detail below.

Dependencies

Version numbers only reflect the versions used for development, other versions may work too. Developed on Linux Mint 18.3 Sylvia.

  • OpenCV 3.4
  • Python 3.5
    • exifread 2.1
    • numpy 1.14
    • opencv-python 3.4
    • rawpy 0.12

extractColor

Extract color values from an image containing a standard x-lite colorchecker grid.

This script attempts to locate color chips in an image by finding areas of max gradient, discarding shapes that don't look like a color chip, and then reconstructing any chips in the grid that it missed. Colors are then averaged within each chip and reported in a csv formatted for use with computeCCM. This auto-detection feature could be a lot more intelligent than it currently is, if anyone wants to implement this the right way!

Currently, only 8-bit png and 10-bit dng source images have been tested, but other formats up to 16-bit should work as well. Example images are provided in the img directory for testing.

Usage

$ extractColor.py [-h] -x X -y Y [-g GAMMA] [-v] input_image output_csv

Required arguments:

  • input_image Source image
  • output_csv Path to save color information
  • -x X Expected width of color chips, in pixels
  • -y Y Expected height of color chips, in pixels

Optional arguments:

  • -h, --help Show help message and exit
  • -g GAMMA, --gamma GAMMA Gamma value of input image, default 1.0 (no gamma correction)
  • -v, --verbose Verbose output

computeCCM

Compute the color correction matrix (CCM) necessary to convert one set of colors to another in XYZ color space.

This script, given two sets of 24x3 sRGB color information A and B, simply converts both to XYZ color space and solves the equation Ax = B, where x is a 3x3 CCM. The same CCM can't perfectly match all 24 colors, so a least squares routine (numpy.lingalg.lstsq()) is used to find the CCM that minimizes overall error.

Example data are provided in the data directory.

Usage

$ computeCCM.py [-h] [-g GAMMA] [-i ILLUMINANT] [-v] reference_csv source_csv output_csv

Required arguments:

  • reference_csv CSV containing reference color information, to be matched
  • source_csv CSV containing source color information, to be converted
  • output_csv Path to save the calculated CCM as a CSV

Optional arguments:

  • -g GAMMA, --gamma GAMMA Gamma value of reference and source data
  • -h, --help Show help message and exit
  • -i ILLUMINANT, --illuminant ILLUMINANT lluminant of source and reference images (default D65)
  • -v, --verbose verbose output

correctColor

Apply color correction and other operations to a raw or processed source image.

This script corrects a given image using a ccm as calculated by computeCCM using the following routine:

  • Normalize (convert to range 0 - 1)
  • Linearize (remove any existing gamma correction, as indicated by the -g flag)
  • Convert to XYZ color space
  • Correct colors by applying the provided CCM
  • Convert back to sRGB color space
  • Apply gamma correction (gamma = 2.2 unless specified)
  • If source image is a raw format, perform auto white balance and black level
  • Optionally apply auto brightness adjustment
  • Save and display result

This is my first attempt at a pipeline for developing raw images, so improvements to any step in this pipeline are welcome.

Example images and data are provided in the img and data directories.

Usage

correctColor.py [-h] [-b] [-g GAMMA] [-i ILLUMINANT] [-v] ccm input [output]

Required arguments:

  • ccm CSV containing the CCM to apply
  • input Source image, processed or raw

Optional arguments:

  • output Path to save the corrected image, if desired
  • -b, --brightness Auto-brightness adjustment (done automatically if necessary)
  • -g GAMMA, --gamma GAMMA Gamma value of source img (default 1, no gamma applied)
  • -h, --help Show help message and exit
  • -i ILLUMINANT, --illuminant ILLUMINANT Illuminant, D50 or D65 (default D65)
  • -v, --verbose Verbose output

Example

This example demonstrates how to extract color info from a source image and a reference image, compute the CCM to convert the source colors to match the reference colors, and then apply that CCM to generate a corrected image.

Commands

./extractColor.py -g2.2 -x35 -y25 img/example_render.png data/example_colors_render.csv

./extractColor.py -g2.2 -x34 -y20 img/example_ref.png data/example_colors_ref.csv

./computeCCM.py data/example_colors_ref.csv data/example_colors_render.csv data/example_ccm.csv

./correctColor.py -g2.2 -b data/example_ccm.csv img/example_render.png img/example_render_corrected.png

Result

Source image to correct:

source image to correct

Reference image to match:

reference image to match

Corrected source image:

corrected image

License

This repo was originally published under the MIT license. It has been heavily modified from its source, but I'm leaving the MIT license as-is.

See the Dependencies section for third-party dependencies, each of which is published under its own license.

References

Original repo: lighttransport/colorcorrectionmatrix