![](bg1.png)
CIRCLE DETECTION USING THE IMAGE RAY TRANSFORM
A Novel Technique for using a Ray Analogy to Extract Circular Features
Alastair H. Cummings, Mark S. Nixon and John N. Carter
School of Electronics and Computer Science, University of Southampton, Southampton, U.K.
Keywords:
Feature extraction, Image processing analogy, Ray-tracing, Wave, Light ray, Transform, Circle detection.
Abstract:
Physical analogies are an exciting paradigm for creating techniques for image feature extraction. A transform
using an analogy to light rays has been developed for the detection of circular and tubular features. It uses a
2D ray tracing algorithm to follow rays through an image, interacting at a low level, to emphasise higher level
features. It has been empirically tested as a pre-processor to aid circle detection with the Hough Transform
and has been shown to provide a clear improvement over standard techniques. The transform was also used on
natural images and we show its ability to highlight circles even in complex scenes. We also show the flexibility
available to the technique through adjustment of parameters.
1 INTRODUCTION
Physical analogies are an exciting paradigm in com-
puter vision that enable the creation of novel tech-
niques that approach the problems of feature ex-
traction from entirely different angles (Nixon et al.,
2009). These analogy based techniques have the ad-
vantage of being based on physical properties of natu-
ral phenomena such as water, heat or force fields and
so are more easily understood by those using them.
In addition to the intuitive nature of the algorithms
the parameters that are used have meanings that are
clear and have real word analogues. Although anal-
ogy operators are heavily based upon a well defined
physical concept, the analogies can be adapted out-
side this definition to increase their effectiveness and
flexibility whilst maintaining the strengths provided
by the analogy. These properties are a clear advantage
over many standard techniques for which the mechan-
ics can be hard to grasp and parameter selection is not
clear.
Heat flow has been used a number of times due to
its smoothing properties. Anisotropic diffusion (Per-
ona and Malik, 1990) is an edge-aware smoothing
technique that allows heat to flow across areas of low
but not high edge strength, so preserving the impor-
tant edge features whilst reducing noise. Direkoglu
and Nixon developed two techniques using a heat flow
analogy for finding moving edges and image segmen-
tation. The first (Direkoglu and Nixon, 2006) used
anisotropic diffusion to smooth images and then heat
flow in the temporal dimension for extraction. The
second (Direkoglu and Nixon, 2007) used heat flow to
segment the image into non-contiguous sections and
then used geometric heat flow to smooth the bound-
aries. This segmentation technique produced encour-
aging results which capitalised on the use of analo-
gies.
Hurley’s force field transform (Hurley et al., 2005)
generates a force field from an image that is analogous
to a gravitational or magnetic field. Each pixel is as-
sumed to attract every other pixel with a force depen-
dent on its intensity and the inverse square law. The
sum of these forces generates a vector field represent-
ing an image. This force field can help in feature ex-
traction and was successfully utilized to create an ear
biometric. Xie et al. (Xie and Mirmehdi, 2008) also
used a force field analogy, generated from mageneto-
static theory and joined with an active contour model
to enable contour detection. In their model the image
border and evolving contour are assumed to have an
electric current running through them and the inter-
action of these currents generates a force field. This
field guides the development of the contour, changing
along with it to guide it to the image border.
This paper describes the creation of a novel trans-
form based on using an analogy to rays of light. The
rules of ray reflection and refraction are employed to
enhance tubular and circular features within an im-
age. The physical rules governing rays are described
in section 2.1 whilst section 2.2 introduces the image
ray transform. Section 3 documents empirical tests on
23
H. Cummings A., S. Nixon M. and N. Carter J. (2010).
CIRCLE DETECTION USING THE IMAGE RAY TRANSFORM - A Novel Technique for using a Ray Analogy to Extract Circular Features.
In Proceedings of the International Conference on Computer Vision Theory and Applications, pages 23-32
DOI: 10.5220/0002818900230032
Copyright
c
SciTePress