Integrating colour correction algorithms

Fang, Fufu (2021) Integrating colour correction algorithms. Doctoral thesis, University of East Anglia.

Preview

PDF Download (22MB) | Preview

Abstract

Digital cameras sense colour different than the human visual system (HVS). Digital cameras sense colour using imaging sensor, whereas the HVS senses colour using the cone photoreceptors in our retina. Each digital camera model has its own device specific spectral sensitivity function. It is therefore necessary to convert the device specific colour responses of an imaging sensor to values that are related to the HVS. This process is typically referred to as colour correction, and it is common to the image processing pipeline across all cameras.

In this thesis, we explore the topic of colour correction for digital cameras. Colour correction algorithms establish the mapping between device specific responses of the camera with HVS related colour responses. Colour correction algorithms typically need to be trained with datasets. During the training process, we adjust the parameters of the colour correction algorithm, in order to minimise the fitting error between the device specific responses and the corresponding HVS responses.

In this thesis, we first show that the choice of the training dataset affects the performance of the colour correction algorithm. Then, we propose to circumvent this problem by considering a reflectance dataset as a set of samples of a much larger reflectance space. We approximate the convex closure of the reflectance dataset in the reflectance space using a hypercube. Finally we integrate over this hypercube in order to calculate a matrix for linear colour correction. By computing the linear colour correction matrix this way, we are able to fill in the gap within a reflectance dataset.

We then expand upon the idea of reflectance space further, by allowing all possible reflectances. We explore an alternative formulation of Maximum Ignorance with Positivity (MIP) colour correction. Our alternative formulation allows us to develop a polynomial variant of the concept. Polynomial MIP colour correction is far more complex thant MIP colour correction in terms of formulation. Our contribution is theoretically interesting, however practically, it delivers poorer performance.

Item Type:	Thesis (Doctoral)
Faculty \ School:	Faculty of Science > School of Computing Sciences
Depositing User:	Chris White
Date Deposited:	23 Feb 2022 10:53
Last Modified:	23 Feb 2022 10:53
URI:	https://ueaeprints.uea.ac.uk/id/eprint/83519
DOI:

Downloads

Actions (login required)

View Item