Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.560603
Title: The image ray transform
Author: Cummings, Alastair
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2012
Availability of Full Text:
Access through EThOS:
Access through Institution:
Abstract:
Image feature extraction is a fundamental area of image processing and computer vision.There are many ways that techniques can be created that extract features and particularly novel techniques can be developed by taking influence from the physical world.This thesis presents the Image Ray Transform (IRT), a technique based upon an analogy to light, using the mechanisms that define how light travels through different media and analogy to optical fibres to extract structural features within an image. Through analogising the image as a transparent medium we can use refraction and reflection to cast many rays inside the image and guide them towards features, transforming the image in order to emphasise tubular and circular structures. The power of the transform for structural feature detection is shown empirically in a number of applications, especially through its ability to highlight curvilinear structures. The IRT is used to enhance the accuracy of circle detection through use as a preprocessor, highlighting circles to a greater extent than conventional edge detection methods. The transform is also shown to be well suited to enrolment for ear biometrics, providing a high detection and recognition rate with PCA, comparable to manual enrolment. Vascular features such as those found in medical images are also shown to be emphasised by the transform, and the IRT is used for detection of the vasculature in retinal fundus images. Extensions to the basic image ray transform allow higher level features to be detected. A method is shown for expressing rays in an invariant form to describe the structures of an object and hence the object itself with a bag-of-visual words model. These ray features provide a complementary description of objects to other patch-based descriptors and have been tested on a number of object categorisation databases. Finally a different analysis of rays is provided that can produce information on both bilateral (reflectional) and rotational symmetry within the image, allowing a deeper understanding of image structure. The IRT is a flexible technique, capable of detecting a range of high and low level image features, and open to further use and extension across a range of applications.
Supervisor: Nixon, Mark Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.560603  DOI: Not available
Keywords: QA75 Electronic computers. Computer science
Share: