Advances in image quality assessment have shown the benefits of modelling functional components of the human visual system in image quality metrics. Visual saliency, a crucial aspect of the human visual system, is increasingly investigated recently. Current applications of visual saliency in image quality metrics are limited by our knowledge on the relation between visual saliency and quality perception. Issues regarding how to simulate and integrate visual saliency in image quality metrics remain. This thesis presents psychophysical experiments and computational models relevant to the perceptually-optimised use of visual saliency in image quality metrics. We first systematically validated the capability of computational saliency in improving image quality metrics. Practical guidance regarding how to select suitable saliency models, which image quality metrics can benefit from saliency integration, and how the added value of saliency depends on image distortion type were provided. To better understand the relation between saliency and image quality, an eye-tracking experiment with a reliable experimental methodology was first designed to obtain ground truth fixation data. Significant findings on the interactions between saliency and visual distortion were then discussed. Based on these findings, a saliency integration approach taking into account the impact of distortion on the saliency deployment was proposed. We also devised an algorithm which adaptively incorporate saliency in image quality metrics based on saliency dispersion. Moreover, we further investigated the plausibility of measuring image quality based on the deviation of saliency induced by distortion. An image quality metric based on measuring saliency deviation was devised. This thesis demonstrates that the added value of saliency in image quality metrics can be optimised by taking into account the interactions between saliency and visual distortion. This thesis also demonstrates that the deviation of fixation deployment due to distortion can be used as a proxy for the prediction of image quality.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:716029 |
Date | January 2017 |
Creators | Zhang, Wei |
Publisher | Cardiff University |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://orca.cf.ac.uk/100239/ |
Page generated in 0.002 seconds