Camera Color Correction Using Two-Dimensional Transform (rethink)

Intorduction

Inpsired by Ref. [1], I try to check how much accuracy a linear 3 $\times$ 3 matrix can produce in color space conversion / color transformation.

The Dataset used

1. Spectral Reflectance Distributions (SRDs) come from Colour Constancy Synthetic Test Data[2].
2. Camera Spectral Sensitivity (CSS) come from Flying Drone Multi-Illuminant Test Set[3].

Code

Prerequisite

1. Python == 3.9.18 or higher (finished and tested on Python 3.9.18)
2. color-sciecne == 0.4.4
3. RUN ColorTransformation2DLUT.ipynb.

Results

1. calculate naive RGB responses (in device-dependent space defined by CSS) at 3200K blackbody radiance (typicall indoor illuminant).
2. calculate ACES RGB responses (Academy Color Encoding Specification - Reference Image Capture Device) at CIE D65.
3. normalize 1. and 2. to the training input ($p$ abd $q$ based on Eq.(3)[1]) and output ($R_{LUT}$, $G_{LUT}$, and $B_{LUT}$ based on Eq.(4)[1]) of the proposed 2D-LUT method[1].
4. generate the visualization (i.e. Figure 6.[1]) to illustrate the residual from linear 3 $\times$ 3 matrix for color space conversion.

Reference

[1] Jon S. McElvain, Walter Gish, "Camera Color Correction Using Two-Dimensional Transforms" in Proc. IS&T 21st Color and Imaging Conf., 2013, pp 250 - 256.
[2] Barnard, K., Martin, L., Funt, B. and Coath, A., A data set for color research. Color Res. Appl., 27, 2002 : 147-151.
[3] Hoda Aghaei, Brian Funt, "A Flying Gray Ball Multi-illuminant Image Dataset for Color Research" in Journal of Imaging Science and Technology, 2020, pp 050411-1 - 050411-8.