Colour Correction using Splines
Suominen, Joni (2024)
2024
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-05-23
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202404203960
https://urn.fi/URN:NBN:fi:tuni-202404203960
Tiivistelmä
The colours of a scene, as recorded by a camera before any processing, differ significantly from those seen by the human eye. The reason is simple: no digital image sensor has the same spectral sensitivities as the human eye. No such sensor can be produced reliably enough due to variations in the manufacturing process. In addition to this, the development of the image sensor involves trade-offs that take precedence, such as minimising noise. The differences in colour reproduction are compensated for in software inside the Image Signal Processor (ISP) during a step commonly known as colour correction.
This thesis presents a method for colour correction involving Penalized B-splines (P-splines). B-splines, being piecewise polynomials with minimal support, are a natural expansion of the polynomial methods previously proposed in the literature. In addition, regularization is included in the proposed model through second-order differences of coefficients to impose smoothness in the solution. This allows for more complex models to be fitted without overfitting and robustness to noise.
In addition, a comprehensive comparison is conducted against the state-of-the-art methods, such as Root-Polynomial Regression and neural networks. For this purpose, an open-source test bench for evaluating colour correction algorithms is published.
The study shows that the method performs comparably or better than the previous methods across different cameras and training datasets, with minimal tuning required. The disadvantage of the proposed model is that it is not inherently invariant to exposure changes. As such, the effects of penalization with regard to exposure and the nature of the transformation are explored.
This thesis presents a method for colour correction involving Penalized B-splines (P-splines). B-splines, being piecewise polynomials with minimal support, are a natural expansion of the polynomial methods previously proposed in the literature. In addition, regularization is included in the proposed model through second-order differences of coefficients to impose smoothness in the solution. This allows for more complex models to be fitted without overfitting and robustness to noise.
In addition, a comprehensive comparison is conducted against the state-of-the-art methods, such as Root-Polynomial Regression and neural networks. For this purpose, an open-source test bench for evaluating colour correction algorithms is published.
The study shows that the method performs comparably or better than the previous methods across different cameras and training datasets, with minimal tuning required. The disadvantage of the proposed model is that it is not inherently invariant to exposure changes. As such, the effects of penalization with regard to exposure and the nature of the transformation are explored.