Color accuracy in single and multi-illuminant mobile phone imaging
Raschke, Ariane (2024)
Raschke, Ariane
2024
Teknis-luonnontieteellinen DI-ohjelma - Master's Programme in Science and Engineering
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-01-08
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202401021023
https://urn.fi/URN:NBN:fi:tuni-202401021023
Tiivistelmä
In the last two decades, mobile phone imaging has become ever more prevalent. Due to this an increase in mobile phone camera quality is an important sales point. One of these quality markers is the correct rendering of colors in the images. Accurate colors come from mainly two processes: white balance and color correction.
In this thesis, we investigate white balance and color correction with the help of spectral information of the scene. Modelling both how color comes to be in the camera as well as in the brain helps us optimize the Image Signal Processor to reproduce accurate colors. For the color accuracy measurements we utilize ColorChecker targets and the C00 color error formula. After investigating white balance and color correction in a single illuminant case, we also look at some cases with two illuminants.
We find that white balance utilizing the spectral profile of the scene illuminant together with color correction utilizing either a Monte Carlo simulation or a normalized matrix division gives us the smallest color errors in single illuminant scenes. For two illuminant scenes, we get better results with our weighted white balance approach than using just a naive Image Signal Processor that handles the whole image like just one illuminant was present. However, the areas where the illuminant mixture is more complicated have still some color cast left.
To get better results for two illuminant scenes the weighting for white balance coefficients needs to be rethought, taking into consideration the intensity of each illuminant separately. With both the intensity and spectral profile of the illuminant, we expect improvements for the model.
In this thesis, we investigate white balance and color correction with the help of spectral information of the scene. Modelling both how color comes to be in the camera as well as in the brain helps us optimize the Image Signal Processor to reproduce accurate colors. For the color accuracy measurements we utilize ColorChecker targets and the C00 color error formula. After investigating white balance and color correction in a single illuminant case, we also look at some cases with two illuminants.
We find that white balance utilizing the spectral profile of the scene illuminant together with color correction utilizing either a Monte Carlo simulation or a normalized matrix division gives us the smallest color errors in single illuminant scenes. For two illuminant scenes, we get better results with our weighted white balance approach than using just a naive Image Signal Processor that handles the whole image like just one illuminant was present. However, the areas where the illuminant mixture is more complicated have still some color cast left.
To get better results for two illuminant scenes the weighting for white balance coefficients needs to be rethought, taking into consideration the intensity of each illuminant separately. With both the intensity and spectral profile of the illuminant, we expect improvements for the model.