junereo/colorcorrectionmatrix

Compute color correction matrix in python and C++

Compute Color Correction Matrix (CCM)

We compute Color Correction Matrix A. In other words, we calculate a 4x3 matrix A which approximate the following equation.

Let P be a reference color checker matrix (24 x 3) and O be a color checker matrix to correct (24 x 3).
P = [O 1] A

Data

We have to prepare color checker patch data as csv format. There are example data in data directory.

• data/colorchart_photo_strobo_linear.csv
• data/colorchart_rendered_strobo_linear.csv

They are 24x3 matrix. The data are made by reading pixel values using Natron2

Dependency

• Python
  • numpy
  • matplotlib
  • Pillow
  • OpenEXR
• C++
  • args.hxx(included in this repo)
  • Eigen3

Build c++ version of computeCCM

$ cd cpp
$ mkdir build
$ cmake ../
$ make

Usage

# computeCCM.py [-h] [-g GAMMA] reference_csv source_csv output_csv
$ computeCCM.py data/colorchart_photo_strobo_linear.csv data/colorchart_rendered_strobo_linear.csv ccm.csv

This command generates optimal Color Correction Matrix as csv file (ccm.csv)

Test

We can compare reference data and corrected data using plotChart.py

$ plotChart.py ccm.csv data/colorchart_photo_strobo_linear.csv data/colorchart_rendered_strobo_linear.csv ccm.csv chart

Each patch shows reference color and corrected color. Upper one is reference and lower one is corrected color. The numbers mean relative error.

Color Correction

Correct given image using CCM. correctColor.py reads jpg or png images, and correctColorExr.py reads exr images.

Usage

$ correctColor.py ccm.csv reference.png corrected

corrected image

Image Diff

Generate diff image between a reference image and a corrected image. We compute a difference between two images and take average of rgb for each pixels.

Usage

$ imageDiff.py photo_reference.png corrected.png

The difference is small as the color approaches blue and the difference is big as the color approaches red.

License

CCM is licensed under MIT license.

Third party licenses

• args.hxx is licensed under MIT License
• Eigen is licensed under Mozilla Public License

References

• RGB coordinates of the Macbeth ColorChecker, Danny Pascale. June 1st, 2006 version. http://www.babelcolor.com/index_htm_files/RGB%20Coordinates%20of%20the%20Macbeth%20ColorChecker.pdf
• Color Correction Matrix http://www.imatest.com/docs/colormatrix/
• Raw-to-raw: Mapping between image sensor color responses. CVPR 2014. https://www.cv-foundation.org/openaccess/content_cvpr_2014/papers/Nguyen_Raw-to-Raw_Mapping_between_2014_CVPR_paper.pdf