Perception-Aware Optimisation Methodologies for Quantum Dot Based Displays and Lighting

Abstract

Human colour vision acuity is limited. This presents opportunities to leverage these perceptual limits to achieve engineering optimisations for devices and systems that interact with the human vision system. This dissertation presents the results of few investigations we carried out into quantifying these limits and several optimisation methodologies that we devised. The first step in this process is to quantify the acuity of human colour vision. We obtained a large corpus of colour matching data from a mobile video game called Specimen. We examine what questions about human vision this dataset allows us to answer and explore global statistics about colour vision based on this data on 41,000 players from 175 countries. We show that we can use the information in this dataset to infer potential candidate functions for the spectral sensitivities of each person in the dataset. The human eye acts like a many to one function; quantifiably different spectra can look like the same colour. This is referred to as metamerism. From a device perspective, different spectra consume different amounts of energy to generate. We show that we can use these two properties to elicit the same colour sensation using less energy. In the colour samples we evaluated, we show that we can achieve up to 10 times less power consumption while achieving a colour match. Given that one cannot change the emission spectrum of a display after fabrication, we propose the use of a multi-primary colour display to achieve this. We present two indices for quantifying the metameric capacity of such a display and its ability to save energy. The emission spectrum of a quantum dot (QD) based device is very narrow. Previous work in the literature suggested that narrow bandwidth spectra can lead to observer metameric breakdown; different observers disagreeing on the perceived ‘colour’ of a spectrum. We show that this might not be the case, using modern colour science tools, and show how metameric breakdown in a display could be minimised by carefully choosing the primary emission wavelengths. The limited colour acuity of human vision implies that people cannot notice small differences in colour. This fact has been used to create approximate colour transformation algorithms that subtly change colours in images such that they consume less energy when displayed on an emissive pixel display without causing unacceptable visual artefacts. We conducted a user study to gather information about the effect of one such colour transform called Crayon. We present a method for effectively picking the optimal transform parameters for Crayon, based on the user study results. The method presented calculates these parameters based on the properties of the image being transformed such that the power saving can be maximised while minimising the loss of image quality. The user study results show that we can achieve up to 50% power saving with a majority of the study participants reporting a negligible degradation in image quality in the transformed images. We additionally investigate a hypothesis that was presented stating that images with large amounts of highly luminous pixels cause increased power consumption in OLED displays due to localised display heating. We show that this hypothesis is wrong. We also investigate if sub-pixel rendering in Pentile displays can be used to reduce display power consumption by intentionally turning off random sub-pixels. However, we present a negative result showing that even single-pixel artefacts are observable on the test platform and thus, this cannot be used to improve display power efficiency. The narrow-band optical emissions of QD based devices mixed with their ability to be fabricated through solution processing can be used to mix multiple QDs together to build devices that generate arbitrary spectral shapes. We show how to use this property in an numerical optimisation based design framework to create lighting devices with a high colour rendering index (CRI). We evaluate the effects of different cost functions and initialisation strategies, and show that, we are able to design devices with a CRI > 96 using only four different QD primaries. We use a charge-transport based simulator to asses the electric properties of the designed devices. We also showcase initial work done on a modular software interface and a material library we developed for this simulator.