Investigations into colour constancy by bridging human and computer colour vision

Abstract

The mechanism of colour constancy within the human visual system has long been of great interest to researchers within the psychophysical and image processing communities. With the maturation of colour imaging techniques for both scientific and artistic applications the importance of colour capture accuracy has consistently increased. Colour offers a great deal more information for the viewer than grayscale imagery, ranging from object detection to food ripeness and health estimation amongst many others. However these tasks rely upon the colour constancy process in order to discount scene illumination to allow these tasks to be carried out. Psychophysical studies have attempted to uncover the inner workings of this mechanism, which would allow it to be reproduced algorithmically. This would allow the development of devices which can eventually capture and perceive colour in the same manner as a human viewer. These two communities have approached this challenge from opposite ends, and as such very different and largely unconnected approaches. This thesis investigates the development of studies and algorithms which bridge the two communities. Utilising findings from psychophysical studies as inspiration to firstly improve an existing image enhancement algorithm. Results are then compared to state of the art methods. Then, using further knowledge, and inspiration, of the human visual system to develop a novel colour constancy approach. This approach attempts to mimic and replicate the mechanism of colour constancy by investigating the use of a physiological colour space and specific scene contents to estimate illumination. Performance of the colour constancy mechanism within the visual system is then also investigated. The performance of the mechanism across different scenes and commonly and uncommonly encountered illuminations is tested. The importance of being able to bridge these two communities, with a successful colour constancy method, is then further illustrated with a case study investigating the human visual perception of the agricultural produce of tomatoes.