Skip navigation
Skip navigation

Spectral image modelling for material-based visualisation

Gu, Lin

Description

This thesis addresses three problems in trichromatic and hyperspectral imagery: outdoor shadow detection, recovery of photometric invariants and personalised image visualisation. It shows how detection of shadows in outdoor imagery is a necessary preprocessing step, before photometric invariants can be separated from a single spectral or RGB image. It then shows how the recovered reflectance parameters can be used to create user colour preferences for visualising the image. The first...[Show more]

dc.contributor.authorGu, Lin
dc.date.accessioned2018-11-22T00:11:35Z
dc.date.available2018-11-22T00:11:35Z
dc.date.copyright2014
dc.identifier.otherb3579013
dc.identifier.urihttp://hdl.handle.net/1885/151763
dc.description.abstractThis thesis addresses three problems in trichromatic and hyperspectral imagery: outdoor shadow detection, recovery of photometric invariants and personalised image visualisation. It shows how detection of shadows in outdoor imagery is a necessary preprocessing step, before photometric invariants can be separated from a single spectral or RGB image. It then shows how the recovered reflectance parameters can be used to create user colour preferences for visualising the image. The first contribution of this thesis is a physics based model to detect shadows in outdoor scenes. Here, we note that shadows arise from diffuse skylight that has been scattered by particles in the atmosphere. This yields a treatment in which shadows in the image can be viewed as a linear combination of scattered sunlight obeying Rayleigh scattering and Mie theory. This allows a ratio to be calculated that permits the problem of recovering the shadowed areas in an image to be recast into a clustering setting making use of active contours. It also allows a metric to be formulated that indicates the degree to which a scene is overcast. Secondly, we address the problem of efficiently recovering the reflectance parameters from a single multispectral or hyperspectral image. To do so, we propose a shapelet based estimator (SBE) to recover shading in an image. The optimisation setting presented here is based on a three-step process. The first calculates the geometric shading based on an alternative shapelets method; the second uses a constrained optimisation approach to update the surface reflectance and the specular coefficients; and the third employs either a simple least-squares formulation or its variants to update the illuminant power spectrum. This yields a computationally efficient method that achieves speed-ups of nearly an order of magnitude as compared to its closest alternative without compromising performance. Moreover, our photometric invariants recovery method can derive the material reflectance and illumination parameters which are necessary for creating user colour preference profiles. To visualise an image, we make use of the heterogeneous nature of a scene and impose consistency over object materials, an approach that allows small compositional variations across objects in the image to be detected. In this way, the quality of the images under consideration can be maximised based on user preferences. This means individual user profiles can be employed for processing real world imagery, while avoiding undesirable effects when the final colour image is produced.
dc.format.extentxx, 161 leaves.
dc.language.isoen_AU
dc.rightsAuthor retains copyright
dc.subject.lcshSpectral imaging
dc.subject.lcshImage processing Digital techniques.
dc.subject.lcshSpectrum analysis Data processing.
dc.subject.lcshPattern recognition systems.
dc.subject.lcshComputer vision
dc.titleSpectral image modelling for material-based visualisation
dc.typeThesis (PhD)
local.contributor.supervisorRobles-Kelly, Antonio
local.description.notesThesis (Ph.D.)--Australian National University
dc.date.issued2014
local.type.statusAccepted Version
local.contributor.affiliationAustralian National University. Research School of Engineering
local.identifier.doi10.25911/5d514d974c440
dc.date.updated2018-11-21T13:18:50Z
dcterms.accessRightsOpen Access
local.mintdoimint
CollectionsOpen Access Theses

Download

File Description SizeFormat Image
b35790131_Gu_L.pdf160.96 MBAdobe PDFThumbnail


Items in Open Research are protected by copyright, with all rights reserved, unless otherwise indicated.

Updated:  19 May 2020/ Responsible Officer:  University Librarian/ Page Contact:  Library Systems & Web Coordinator