The image ray transform for structural feature detection


Cummings, Alastair H., Nixon, Mark S. and Carter, John N. (2011) The image ray transform for structural feature detection Pattern Recognition Letters, 32, (15), pp. 2053-2060. (doi:10.1016/j.patrec.2011.08.020).

Download

[img] PDF ahcprl.pdf - Accepted Manuscript
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (1MB)
[img] PDF __userfiles.soton.ac.uk_Users_nsc_mydesktop_272778carter.pdf - Version of Record
Restricted to Repository staff only

Download (1MB)

Description/Abstract

The use of analogies to physical phenomena is an exciting paradigm in computer vision that allows unorthodox approaches to feature extraction, creating new techniques with unique properties. A technique known as the ``image ray transform'' has been developed based upon an analogy to the propagation of light as rays. The transform analogises an image to a set of glass blocks with refractive index linked to pixel properties and then casts a large number of rays through the image. The course of these rays is accumulated into an output image. The technique can successfully extract tubular and circular features and we show successful circle detection, ear biometrics and retinal vessel extraction. The transform has also been extended through the use of multiple rays arranged as a beam to increase robustness to noise, and we show quantitative results for fully automatic ear recognition, achieving 95.2% rank one recognition across 63 subjects.

Item Type: Article
Digital Object Identifier (DOI): doi:10.1016/j.patrec.2011.08.020
Keywords: vision, feature extraction, circle detection, biometrics, medical imaging
Organisations: Vision, Learning and Control
ePrint ID: 272778
Date :
Date Event
10 September 2011e-pub ahead of print
1 November 2011Published
Date Deposited: 16 Sep 2011 09:57
Last Modified: 17 Apr 2017 17:39
Further Information:Google Scholar
URI: http://eprints.soton.ac.uk/id/eprint/272778

Actions (login required)

View Item View Item