Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.753908
Title: New techniques in derivative domain image fusion and their applications
Author: Hayes, Alex
ISNI:       0000 0004 7426 9934
Awarding Body: University of East Anglia
Current Institution: University of East Anglia
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
There are many applications where multiple images are fused to form a single summary greyscale or colour output, including computational photography (e.g. RGB-NIR), diffusion tensor imaging (medical), and remote sensing. Often, and intuitively, image fusion is carried out in the derivative domain (based on image gradients). In this thesis, we propose new derivative domain image fusion methods and metrics, and carry out experiments on a range of image fusion applications. After reviewing previous relevant methods in derivative domain image fusion, we make several new contributions. We present new applications for the Spectral Edge image fusion method, in thermal image fusion (using a FLIR smartphone accessory) and near-infrared image fusion (using an integrated visible and near-infrared sensor). We propose extensions of standard objective image fusion quality metrics for M to N channel image fusion measuring image fusion performance is an unsolved problem. Finally, and most importantly, we propose new methods in image fusion, which give improved results compared to previous methods (based on metric and subjective comparisons): we propose an iterative extension to the Spectral Edge image fusion method, producing improved detail transfer and colour vividness, and we propose a new derivative domain image fusion method, based on finding a local linear combination of input images to produce an output image with optimum gradient detail, without artefacts - this mapping can be calculated by finding the principal characteristic vector of the outer product of the Jacobian matrix of image derivatives, or by solving a least-squares regression (with regularization) to the target gradients calculated by the Spectral Edge theorem. We then use our new image fusion method on a range of image fusion applications, producing state of the art image fusion results with the potential for real-time performance.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.753908  DOI: Not available
Share: