Author: Damian Richard Tohl
Tohl, Damian Richard, 2019 Image Enhancement and its Applications, Flinders University, College of Science and Engineering
Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.
Images and videos continue to play an important role in daily life, covering a scope that ranges from preserving memories all the way to entertainment and with the advent of technology are becoming more prevalent than ever. Coupled with this, is the desire for better quality, for which image enhancement is a useful tool that can be used to manipulate images and video in order to improve their perceived quality. Likewise, for processes that rely on an image or video source, such as analysis and identification, a good quality input cannot always be guaranteed and so enhancement is used as an important pre-processing step to improve the quality of an input image or video.
Enhancement techniques can improve the perceived quality of images, but many of them introduce other problems which can adversely affect the quality of images. These problems manifest in the form of over- and under-enhancement, a loss of details, halo effects, reduced signal to noise ratio and flickering in video enhancement. The aims of this thesis are to develop enhancement techniques that significantly improve the quality of images and video whilst avoiding the common problems often associated with other enhancement techniques.
One of the techniques we present is a multi-level histogram shape segmentation method that can avoid over and under enhancement by segmenting the histogram in such a way that regions with a similar frequency of occurrence are grouped together and independently equalised.
To achieve brightness preservation for any image independent of the degree of enhancement, we propose a novel method to find a unique S-shaped curve transfer function for each image using successive approximation. As the S-shaped curve reduces intensities below a certain point and increases intensities above this same point, then for each image there exists a unique location of this point that will maintain the brightness of the image. We also propose a modified gamma curve which approximates the S-shaped curve, but has a wider control over the degree of enhancement and greater computational efficiency.
By making use of recently developed image quality measures, which have improved the correlation between human visual perception and the value they produce, a novel optimisation method is proposed that extends the successive approximation approach to maximise any chosen image quality measure value. Not only does this optimise enhancement, but can be used to assess which features of the image a particular image quality measure targets and how well it correlates with visual perception.
A novel adaptive method for detail enhancement is also proposed in this thesis that can suppress noise in homogeneous regions and avoid halo effects while providing detail enhancement to improve the signal to noise ratio.
Results show that the enhancement techniques presented in this thesis provide a desirable degree of enhancement while outperforming other benchmarking algorithms both quantitatively and visually. These novel image enhancement techniques can also be applied practically in order to improve other processes, such as image expansion, whereby the modified gamma curve is used with an adaptive gamma value based on the difference between two different interpolation methods to regain the sharpness of edges in the expanded image. Another process that can benefit from the use of these image enhancement techniques is enhancement prior to colour filter array demosaicking, which can give better results than the current methods which apply enhancement after demosaicking.
Keywords: Image and video enhancement, contrast enhancement, histogram shape segmentation, histogram equalization, brightness preservation, S-shaped curve, modified gamma curve, successive approximation, image quality measure, optimised image enhancement, detail separation, guided image filter, adaptive detail enhancement, image enhancement.
Subject: Computer Science thesis
Thesis type: Doctor of Philosophy
Completed: 2019
School: College of Science and Engineering
Supervisor: Dr Jimmy Li