Colour for the Advancement of Deep Learning in Computer Vision

This thesis explores several research areas for Deep Learning related to computer vision concerning colours. First, this thesis considers one of the most long standing challenges that has remained for Deep Learning which is, how can Deep Learning algorithms learn successfully without using human annotated data? To that end, this thesis examines using colours in images to learn meaningful representations of vision as a substitute for learning from hand-annotated data. Second, is another related topic to the previous, which is the application of Deep Learning to automate the complex graphics task of image colourisation, which is the process of adding colours to black and white images. Third, this thesis explores colour spaces and how the representations of colours in images affect the performance in Deep Learning models.