Sie sind auf Seite 1von 9

Tristage Hough transform for multiple ellipse

extraction

H.K. Muammar, MSc


M. Nixon, PhD, MlEE

Indexing terms: Computer graphics, Design

such as Freeman chain coding and polygonal approx-


Abstract: The importance of rapid and accurate imation (see Reference 3); however, the reliability of these
extraction of ellipses in images for model-based techniques is likely to suffer if the edge data is noisy or if
computer vision systems has long been recognised. the image feature space is cluttered.
The Hough transform technique has established The Hough transform (HT) was introduced in 1962 as
itself as a robust method for shape extraction, a technique for detecting lines in digitised images. Since
however, its application to ellipses has been then, it has been improved and extended to detect other
restricted by excessive computational and storage parametrically defined shapes and arbitrary shapes [4].
requirements. Direct extension of the Hough tech- Its popularity has increased steadily owing mainly to its
nique to ellipses is particularly demanding since a robustness and ability to cope with image noise. In its
five-dimensional accumulator array is required. standard form, the H T has been shown to be equivalent
This paper presents a novel tristage technique for to template matching [SI, and as such, maximises the
ellipse extraction based on decomposing the signal-to-noise ratio of shape detection when white noise
problem into sequentially executed stages. Candi- is present. The process of gathering evidence in param-
date ellipse centres are determined using an eter space independently from edge data in the image
improved centre-finding procedure and novel space, indicates that the H T is capable of recognising
Hough-based procedures extract the remaining objects even when they are partially hidden or data seg-
parameters in two further stages. Although multi- ments are missing.
stage parameter extraction is not new, many of the A number of disadvantages are associated with the
problems that exist in previous schemes are standard implementation of the Hough transform for
resolved. In the new technique, the dimensionality ellipses. An ellipse can be conveniently parameterised by
of the accumulator arrays used does not exceed its centre co-ordinates, major and minor radii and orien-
two, while their ranges are predefined. Memory tation. An extension of the standard parametric H T to
savings are therefore significant. Concentric ellipses imposes the requirement of a five-dimensional
ellipses of the same orientation are shown to be parameter space. If accurate extraction of ellipses is to be
extracted with little extra computational demand. carried out, a uniformly fine quantisation of this param-
Two extensions of the tristage technique to multi- eter space is necessary. This, however, can lead to unac-
ple ellipse extraction are presented and compared. ceptable demands in terms of computer storage required
When applied to real images, the new technique and the computation needed to perform the H T
extracted multiple, concentric ellipses reliably and mapping. The task of extracting ellipses from images can
with reduced complexity. be made practical by making use of the gradient direction
of contour points and by breaking down the problem
into one that can be solved in two, or more, sequentially
1 Introduction executed stages. A number of ellipse recognition schemes
that have followed this approach have emerged.
I. 1 Ellipse detection
A fundamental requirement in many computer vision I .2 Previous approaches
systems is the ability to extract parametrically defined Tsuji and Matsumoto [6] have presented a scheme for
shapes from images. Many manufactured objects are cir- ellipse recognition whereby the centres of candidate
cular, and appear as ellipses when viewed from an ellipses are extracted during the initial pass. The tech-
oblique angle. Two-dimensional contour information can nique by which ellipse centres are found is based on a
be used in three-dimensional shape analysis, to improve simple geometric property of ellipses. If a pair of points
the accuracy and robustness of object-shape matching with equal tangents lie on the contour of the same ellipse,
[l]. In ACRONYM [2], ellipse contour data is used to then their midpoint is the centre of the ellipse. In the first
aid the process of object recognition. The rapid and accu- stage of the centre-finding procedure, pairs of image
rate extraction of ellipses from images can, therefore, be points with equal, or near equal, tangents are extracted.
highly important for any model-based vision system. Fast A two-dimensional histogram of the midpoints of each
extraction of ellipse contours is possible using methods extracted pair of points is then constructed. Peaks in this
histogram define centres of candidate ellipses in the
Paper 7412E (C2), first received 22nd September 1989 and in revised
form 20th March 1990 image. The set of points which contributed to the
The authors are with the Department of Electronics and Computer Sci- extracted centre are then identified and a least mean
ence, University of Southampton, Highfield, Southampton, SO9 5NH, squares technique is applied to these points to obtain
United Kingdom estimates of all five parameters. The least squares method
IEE PROCEEDINGS-E, Vol. 138, N O . 1, J A N U A R Y 1991 21
can, however, suffer if outliers or irrelevant data is (AHT) [lo] is used to extract the remaining three param-
present in the image. Such a situation is likely to arise if eters from a parameterisation based on the polynomial
concentric ellipses exist in the image. To identify the exis- representation of the ellipse (eqn. 1). The AHT implemen-
tence of concentric ellipses a geometric test is applied. tation uses a small (9 x 9 x 9) accumulator array and
The least squares technique is then applied separately to iteratively focusses in on parameter regions of highest
each ellipse in the concentric set, if it exists. The centre- accumulator density. It does this by dynamically and
finding technique they use has several limitations. If, independently increasing the resolution of each param-
owing to occlusion, insufficient pairs of points with equal eter in the array, thereby reducing its coverage. It over-
tangents remain on the ellipse contour, the technique is comes the main problem associated with the polynomial
liable to fail. Secondly, straight lines in the image must be parameterisation of the ellipse, namely that of large
located and removed before ellipses are searched for, as storage requirements owing to large parameter ranges.
these generate false ridges during the centre-finding However, problems with the AHT include its complex
process. implementation. Thin, ridge-like peaks in parameter
Tsukune and Goto [7] apply the same centre-finding space can result in errors during the redimensioning stage
technique, but recover the remaining parameters using an of the parameter space and this can result in incorrect
implementation based on the polynomial representation peak identification.
of the ellipse. If a translation is applied to image points
such that their origin lies on the estimated ellipse centre, 1.3 The tristage approach
the polynomial equation of an ellipse can be expressed as In this paper, a novel scheme for ellipse recognition is
follows: presented whereby ellipse parameters are extracted in
three sequentially executed stages. During the first pass,
x2 + By2 + 2Dxy + C = 0 B - D2 > 0 (1) candidate ellipse centres are found using the centre-
Extracting B, C and D using a direct Hough transform- finding method adopted by Yuen et al. [9]. The remain-
ation based on this parameterisation would require a ing parameters are found in two further stages using
three-dimensional accumulator array. Tsukune and Goto Hough-based parameterisations that overcome the prob-
decompose this problem into two further steps by using lems associated with previous schemes. Once a candidate
an alternative parameterisation obtained by differentiat- ellipse centre has been determined, a set of points lying
ing eqn. 1 with respect to x . This gives on the axes of symmetry of the ellipse is generated. These
points are then mapped to a one-dimensional orientation
+ + +
x Byy' D(xY' y ) = 0 (2) histogram from which ellipse orientation is determined.
Using edge gradient direction y', a two-dimensional HT Ellipse contour points are then passed to a third stage
is applied to determine B and D. Once B and D have where the estimated centre and orientation are used to
been estimated, a one-dimensional histogram can be used determine the major and minor ellipse radii. Pairs of
to estimate the final parameter C using eqn. 1. A contour points are rotated through an angle equal to the
number of disadvantages exist with this scheme. The measured orientation about the estimated ellipse centre.
range of parameters used in the polynomial representa- Ellipse radii are subsequently found by applying a novel
tion of an ellipse is generally large and difficult to define. two-dimensional HT to the rotated points. A block
This can mean that large amounts of computer storage diagram of the tristage technique is shown in Fig. 1.
are required, even for the two-dimensional accumulator Errors in edge direction estimates are less likely to be

+
array. Savings in computer storage may therefore not be introduced during the estimation of orientation and
significant. Secondly, edge gradient direction information
is used extensively throughout the scheme. Errors as a acquire image
result of inaccurate edge direction estimation are likely to
propagate to the second stage of the ellipse recognition
procedure. This can result in inaccurate estimation of edge detecton and
orientation and radii of the ellipse. gradient direction
estimation
Nagata, Tamura and Ishibashi [SI describe a recursive
least mean squares based scheme for detecting ellipses.
Before applying the least squares technique, the image is
segmented using a procedure that is based on statistically
determining an optimum set of thresholds from a multi-
mode histogram of the image. This is followed by the _ _ _
I
I
application of a noise reduction method to the binary centre finding I
I
image. Region boundaries are finally obtained by apply- I
I feature pant labelling
ing a contour tracking algorithm to each region. The I I
I
recursive least squares method is applied to the extracted I
contours to recover all five parameters. As with most
least squares procedures, the presence of irrelevant points
can be detrimental to the accuracy of the technique. I
r
v
-find Orientation
I
use symmetry
of ellipse
I
I
Careful consideration must be given to the choice of an I
appropriate weighting scheme which ensures that outlier rotate points an3
points d o not significantly affect the overall calculation of apply HT
I
parameters.
A new centre-finding technique has recently been pre-
sented by Yuen, Illingworth and Kittler [9] as part of a
scheme for ellipse detection. Their improved centre-
finding technique can tolerate higher levels of partial
occlusion of the ellipse. The adaptive Hough transform Fig. 1 Tristage Hough transform for ellipse recognition

28 I E E PROCEEDINGS-E, Vol. 138, N o . I , J A N U A R Y 1991


ellipse radii, since edge positional information is used The co-ordinates of T are given as:
mainly in the later stages. The problem of undefined
parameter ranges encountered in the polynomial param- t, =
Y , - Yz - SlXl + 92x2
eterisation of an ellipse is resolved. Parameter limits in 92 - 91
the tristage technique bear a one-to-one relationship with
D the possible parameter values the unknown ellipse may t, =
Yl92 - Y291 + 9 2 9 1 ( x 2 - x1)
take. Consequently, significant savings in computer - 91
g2
storage are achieved. The extraction of concentric ellipses
with equal orientation can be carried out efficiently since and for the point M :
these can be accurately resolved by the algorithm in the
final stage. Previous methods based on the polynomial mx = ~
+x2 Yl
=~ + Y2
parameterisation of an ellipse [7,9] are not directly 2 2
suited to the extraction of concentric ellipses. Least
where g1 and g2 are the slopes of the tangents of P , and
squares based techniques [6, 81 require additional tests to
determine the existence of concentric ellipses. P , , respectively.
Lines constructed from all pairs of points lying on the
same ellipse will intersect at one point, O ( x , , yo), the
2 Details of the tristage technique centre of the ellipse. In a practical implementation of this
2.1 Centre estimation scheme, P , and P , cast their votes in a two-dimensional
The first stage in most Hough-based ellipse detection accumulator array called the centre array. Locations in
schemes is that of centre-finding. It is perhaps the most this array which intersect the line T M are incremented.
important, since the accuracy with which the remaining Local peaks in the array give evidence for possible
three parameters can be determined frequently depends ellipses and their centres in the image. Votes cast by pairs
on the ellipse centre measurements. Successful extraction of points resulting from random noise or other shapes in
of the ellipse boundary is made possible if good centre- the image will, in most cases, give rise to background
finding techniques are used. One centre-finding method accumulation in the centre array. In certain situations
(used in References 6 and 7) evaluated the midpoints of however, the accumulation of votes caused by pairing
pairs of contour points with approximately equal points on different ellipses or shapes can result in false
tangents. If the contour points were both located on the peaks in the accumulator array. These peaks may disrupt
boundary of an ellipse then their midpoint defined the the interpretation of the parameter space by masking out
ellipse centre. A two-dimensional histogram was used to significant peaks relating to the centres of smaller or par-
accumulate the midpoint locations. Peaks in the histo- tially occluded ellipses.
gram defined the locations of possible ellipse centres. If the maximum expected size of ellipse is taken into
However, partial occlusion of the ellipse, often resulted in account, then the likelihood of false peaks arising can be
a significant reduction in the number of points symmetri- reduced by setting upper limits on the spacing of pairs of
cal about the ellipse centre. In these situations, the tech- points which are included in the voting process. Pairs of
nique was liable to fail. points whose separation exceeds this limit are assumed
Because of its ability to handle partial occlusion, we unlikely to lie on the contour of the same ellipse. Simi-
have used the ellipse centre-finding method of Yuen et al. larly, we can avoid pairing points which are too close to
[9]. This procedure is based on a simple geometric pro- each other, as errors in edge gradient direction estimation
perty of ellipses. Consider the two points, p , ( ~ ,y,), and are likely to result in an invalid set of votes being gener-
P 2 ( x 2 ,y2), lying on the boundary of an ellipse as shown ated.
in Fig. 2. Given that P , and P , have intersecting Hence, we need only include a pair of image points,
P , ( x , , y,) and P 2 ( x , , y 2 ) , in the voting process if they
satisfy the following criteria:

where d,,, and dminare some preset values. This criterion


was applied by Yuen et al. [9] in their implementation of
the ellipse centre-finding scheme. Careful choice of d,,,
and dminprovided a degree of control over the particular
size of ellipse that was extracted. In general, d,,, was set
to a value equal to twice the expected maximum of the
major radius.
The process of pairing up a set of points results in a
Fig. 2 Ellipse showing intersecting tangent point T , and midpoint M
line along T M that passes at some point through the
centre of the ellipse. For practical purposes, lines of
tangents, then the line through the intersection point length L, starting from M , in the direction of the expected
T ( t , , t,), and the midpoint M ( m x , my), of P , and P 2 will ellipse centre were generated. In order that sufficient
pass through the centre of that ellipse. The equation of votes were generated to allow accurate estimates of
line T M can be expressed as : ellipse centres, the length L was set in the region of the
expected maximum of the major ellipse radius.
y=cx+d
Since the Hough transform is an incremental evidence
where x and y are the co-ordinates of points on the line. gathering procedure, shape extraction is possible even
The gradient of the line c , and the offset d are given by: if part of the data is only processed. Point hopping
was introduced to speed extraction, whereby qualifying
t, - my feature points were paired with every nth point in the
C = d = m y - cm,
tx - mx feature list rather than with every point.
IEE PROCEEDINGS-E, Vol. 138, N O . 1, J A N U A R Y 1991 29
2.2 Determining ellipse orientation 2.3 Estimating the major and minor ellipse radii
Once an estimate of the location of an ellipse centre has Once a maximum, O,,,, has been extracted from the
been made, the process of determining ellipse orientation orientation histogram, the estimation of ellipse radii can
can be carried out. If a shape is known to have an axis of proceed. Pairs of points, P,P,, in a third pass through
symmetry with slope 0,, then the angular slopes (in the image are rotated to P;(x;,y;) and P;(x;,y;),
radians) of its k axes of symmetry are given by : through an angle eoptabout the estimated ellipse centre,
Oi = 8, + (i - l)n/k i = 1, 2, . . . , k according to the transformation [123 :
An ellipse possesses two axes of symmetry, 0, and e,, x' = ( x - x o )cos O,,, - ( y - y o ) sin Oopt + xo
whose angular slopes are given by: Y' = ( x - xo) sin 0,,, + (Y - Y o ) cos Oopt + Y o
e, = 0, This results in a subset of transformed points situated on
and the boundary of an unrotated ellipse. For a pair of points
6 2 = eo + 4 2 P',(x;, y ; ) and P 2 ( x ; ,y;) on the contour of an unrotated
The procedure for estimating ellipse orientation involves ellipse, with centre O ( x o ,yo),we have
identifying the set of midpoints M , and tangent intersec-
tions T , which lie on either axis of symmetry of the (3)
ellipse. An important assumption is that the intersection
point of the axes of symmetry of the ellipse lies on the (4- X O ) , (Y; - Yo),
(4)
estimated ellipse centre. If the distances O P , and O P , , + 7 = l
from any given pair of points, P , P , , respectively, to the
el!ipse centre are equal, then the points M and T are These can be combined by eliminating b to give
assumed to lie on either axis of symmetry of the ellipse.
Points M and T which lie on the axes of symmetry are
then rotated by a factor of n/2 about the estimated ellipse
centre, so that they lie in the first quadrant of a co- Solving for a,
ordinate frame whose origin lies on that centre.
The angles subtended by O M and OT increment a
one-dimensional orientation histogram. The mode in the
histogram defines the angular slope of either the major or
the minor axis (0, or 0,) of the ellipse. Which axis the We can also combine the two equations so as to elimi-
angular slope relates to is unimportant at this stage and nate a and solve for b to give
can be resolved after the major and minor radii have
been determined. The procedure for determining ellipse
orientation is outlined in Fig. 3. A weighting scheme
similar to that used by Van Veen and Groen [111 was For a pair of points attributed to lie on the ellipse, eqns.
incorporated into the voting process. The coefficients of a 5 and 6 can be used to calculate values for the major and
Gaussian function whose mean coincided with the angle minor radii, a and b.
subtended by O M and OT were added to the histogram Points P', and P ; vote for major and minor radii in a
locations during the voting process. The Gaussian func- quantised 2-dimensional radius accumulator array. The
tion was chosen to compensate for errors in ellipse orien- location of a peak in this array defines the a and b radii
tation owing to inaccurate edge tangent direction data. of the detected ellipse. Ellipse orientation is generally
The coefficients of the function were evaluated over nine defined as the angle subtended by the major axis with the
elements with a standard deviation (namely width) equal horizontal co-ordinate of the image frame. Symmetry of
to 1.0. This configuration was observed to improve the the ellipse about n/2 means that the ambiguity arising in
robustness of orientation estimates with a negligible the measurement of orientation can be resolved by
increase in computational demand. inspection of the measured a and b radii. If b is greater
DOi=l,M !Pointpi than a, the ellipse orientation becomes
D O j = i + 1, M !Point Pj
dl = O P i
8, = o,,, + n/2
d2 = OPj and a and b are exchanged. Otherwise, the measured
parameters remain valid. In a situation where contours of
IF ( ( d l . E O d 2 ) . A N D (Pi . N E . P i ) ) T H E N
concentric ellipses with equal orientation exist in the
{Work out M and r ) feature map, these will map to separate peaks in the
{Rotate M and T about 0 by a factor of pi/2 so that they lie in radius array.
the first quadrant of a co-ordinate frame centered at 0)
thetal =arctan ( ( m y - y o ) / ( m x - , Y O ) ) 2.4 Feature point labelling and accumulator array
theta2 = arctan ( ( t y - yo)/(tw - , Y O ) ) simplification
acc (thetal ) = acc (thetal ) + 1 Applying the orientation and radius finding procedures
acc (theta2) = acc (theta2) + 1 (stages 2 and 3) directly to a feature map in which candi-
ENDlF date ellipse centres have been extracted is likely to be
computationally expensive. Also, it may yield inaccurate
ENDDO
ENDDO
results if the feature space is complex and multiple
ellipses exist. To reduce the complexity of the image
{The location with maximum votes in array 'acc' is obtained} feature space, it was necessary to identify the points from
Fig. 3 Outline of procedure for determining the axes of symmetry of which the significant accumulator counts originated.
ellipse and orientation That is, a subset of points lying on the contour of the
30 IEE P R O C E E D I N G S - E , Vol. 138, N o . I , J A N U A R Y l Y Y 1
ellipse whose c e n t r e has bcen found was required. Appli- candidate ellipse centres have been accumulated. Each
cation of stages 2 and 3 to the subset of contour points contour point Pi is paired with all points Pi,provided
representing the ellipse boundary meant that procedures that Pi and P j do not have equal tangent gradients. The
implemented for extracting the remaining three param- process of backmapping identifies the accumulator loca-
eters were more robust and less computationally inten- tion aopr,i . Centre array simplification can be carried out
sive. Extraction of the relevant boundary points was by accumulating votes at locations in a second
achieved using a method derived from a backmapping accumulator array of similar dimensions to the centre
technique introduced by Gerig and Klein [13]. The back- array. The process of feature point labelling using the
mapping technique is based on the assumption that a Gerig and Klein procedure reduces the centre-finding
given boundary point should be a member of only one transform to a many-to-one mapping. The identification
curve in image space. Membership to that curve is and extraction of a subset of points labelled as having
defined by an accompanying accumulator cell with the contributed most to the candidate ellipse centre during
highest count, and hence, an optimum parameter set aopr. the centre-finding process is now possible.
To identify the optimum parameter set for a given Yuen et al. [9] have included the backmapping tech-
boundary point, the Hough procedure is repeated a nique of Gerig and Klein [131 in their scheme for ellipse
second time. But instead of incrementing cells intersected recognition. The computational cost of the backmapping
by each voting surface, the location of the cell with the technique, in their implementation is the same as the
highest count a,,,, is found. The boundary point is then centre-finding. The process of centre-finding and back-
labelled with this location. The identification of boundary mapping can be computationally intensive, particularly if
points that contributed to a peak (optimum parameter a large number of points exist in the image. Positional
set) in the accumulator array becomes straightforward. information relating to local peaks in the centre array
Incorporating the Gerig and Klein backmapping pro- was used, in our scheme, to carry out feature point label-
cedure can also result in a simplification of the accumula- ling and centre array simplification in the initial centre-
tor array. During the backmapping procedure, if all cells finding pass. This resulted in an additional saving in
except for aoprthat intersect the voting surface are can- computation.
celled, then the distribution of votes in the accumulator
array is concentrated at peak locations (optimum param- 3 Extending the technique t o detect multiple
eter sets). Sidelobes associated with local peaks generated ellipses
during the voting process can obscure peaks relating to
nearby or smaller ellipses. Accumulator simplification Two extensions of the tristage Hough transform to detect
will suppress these sidelobes thus simplifying the process instances of multiple ellipses are outlined.
of multiple ellipse detection.
3.1 Ellipse contour removal (ECR)
The basic strategy for incorporating the Gerig and
In the first extension, ellipses extracted using the tristage
Klein procedure into the ellipse centre-finding routine is technique are removed from the binary contour map.
shown in Fig. 4. A modified version of the centre-finding
This is carried out by setting edge pixels which lie on the
procedure is applied to the image contour points, after
contour of a simulated ellipse with the extracted param-
c for all eters to zero. The centre-finding procedure is reapplied
and further peaks searched for until the extracted peak is

- i'
below a predetermined threshold. At this point, no
further ellipses are assumed to be present in the image
for all apply the same point and the search ends. Peak threshold selection is arbitrary
pairing criteria as in but is largely governed by the expected ellipse size and

7
generate voting function
f@,,,a)=O
centre finding the extent to which it may be obscured.
3.2 Accumulator peak removal (APR)
Rather than delete the detected ellipse from the binary

I
identify the cell w i t h maximum
votes along f(p,,, a ) = 0,
edge map, we can instead delete the associated peak it1
the accumulator array. This forms the basis of the second
extension. Centre-finding and feature point labelling are
applied only once to points in the binary edge map.
Peaks associated with the extracted ellipse are removed
by setting accumulator cells in a 5 x 5 neighbourhood
centred at the extracted peak location to zero. Remaining
peaks are subsequently searched for until the extracted
accumulate a 1 -dimensional peak is below a predetermined threshold, at which point
- histogram w i t h aoDt,, the search ends. This extension to multiple ellipse detec-
L,(aop,,,I = L , b O D t,,I + 1 tion is more likely to be successful if applied to the sim-

1
plified centre array resulting from the Gerig and Klein
procedure [lo]. This is because, in the simplified centre
array, sidelobes which would otherwise obscure further
identify the cell in L, w i t h peaks resulting from smaller or partially obscured ellipses
maximum votes are suppressed.
aODr I

3.3 Analytic comparison of the complexity of each


1 method
A comparison of the approximate complexity for each
Fig. 4 Flow diayram for Geriy and Klein hackmappiny procedure method can be made by considering the simple case of an
applied to ellipse centre-finding image containing k ellipses each composed of n points.
IEE PROCEEDINGS-E. Vol. 138. N o I , J A N U A R Y 1991 31
Only the computation required during centre-finding will than or equal to 50 pixels could therefore be accommo-
be considered. The number of possible point pairs on one dated.
ellipse is given as Image PARTS, (Fig. 5), is a typical industrial scene
n! and consists of two metal rings, a metal bracket and 9
np bolt, bar;klit by il JiTcueeiJ 1iCjht BOUTSS. A G-annT
2 - - n(n - 1)
(n - 2)! operator [14] was applied to the image and the resulting
feature map, to which the ellipse finding procedures were
In order that accurate feature point labelling can occur
applied, is shown in Fig. 6. Two sets of concentric ellipses
during centre-finding, permutations of points on the
were formed by the metal rings, the rightmost set labelled
ellipse must be considered. Assuming that no point
as concentric set A, the leftmost as concentric set B. The
pairing occurs between different ellipses, then the number
circular cutting in the metal bracket viewed from an
of votes, V,,, , cast by all k ellipses is
angle has resulted in a single ellipse contour. A total of
k
1247 points exist in the feature map of image PARTS. To
Yo* = cv
i= 1
ensure that accurate centre-finding and feature point
labelling would result, d,,, and dminwere set at 60 and 5
where V is the number of votes cast by each ellipse and is pixels, respectively, and L was set at 40 pixels. These are
equal to Ln(n - 1). L is defined as the length, in pixels, of rough figures and were obtained as described earlier.
the voting line cast by each pair of points. The total Pairing each feature point with every 15th in the feature
number of votes, VEcR,cast by all k ellipses using ellipse
contour removal (ECR) is
k-1 k-i

Using the extension based on accumulator peak removal


(APR), and assuming that the cost incurred in peak delet-
ing is negligible in comparison, then the total votes cast is
1

v,,, = 2v
i= 1

The ratio of votes cast by the two extensions can be


expressed as

i= 1

which simplifies to

It appears, therefore, that a saving in the number of votes Fig. 5 Test image P A R T S
cast using the APR extension is only obtained when there
are two or more ellipses in the image. Although this may
be true for the simulation described above, in practice,
real images are often cluttered with irrelevant shapes
resulting in extensive pairing between points other than
those lying on the ellipse contours. It is faster, in practical
situations to delete the peaks associated with the
extracted ellipse than to reaccumulate the centre space.

4 Experimental results
n
In this Section the proposed ellipse contour removal
(ECR) and the accumulator peak removal (APR) exten-
sions of the tristage technique to multiple ellipse detec-
tion were compared experimentally. The algorithms,
coded in VAX FORTRAN 77 and run on a MicroVax 11,
were applied to real images quantised to 256 x 256
pixels, containing multiple ellipses. Storage requirements
for the tristage method were as follows. Ellipse centres
were extracted to integer resolution from a 256 x 256 cell
accumulator array. Orientation and major and minor
radius estimates were obtained to integer resolution from
a 91 element orientation-histogram and a 51 x 51
element radius-accumulator array. Ellipses with arbitrary
orientation and whose major and minor radii were less Fig. 6 Feature map

32 IEE P R O C E E D I N G S - E , Vol. 138, N O . I , J A N U A R Y 1991


map coIisiderably reduced thc computation required by and minor radii of each ellipse in the concentric set were
the centre-finding procedure without noticeably impair- identified simply by extracting peaks in the array which
ing its robustness. exceeded a preset threshold. Identification of ellipse radii
The subsels of kature points identified with the ellipst- i n a cvncentric set is likely to be more complex in
centres are shown in Figs. 7-9. Note that the point label- methods which are based on the polynomial representa-
ling procedure has failed to identify all the points on each tion of the ellipse [7,9]. The existence of concentric
ellipse contour. In addition, points not on the ellipse ellipses would have to be confirmed and additional pro-
contour have also been identified. Outliers present in the cessing required before the least squares technique is
subset of points identified by the feature point labelling applied to the subset of points.
procedure can present problems to l e a t squares based A benchmark for comparing the accuracy of ellipse
methods. These inaccuracies in point labelling have not parameters extracted using the ECR and APR extensions
affected the performance of stages 2 and 3 of the ellipse was provided by a partial implementation of the stan-
finding procedure. dard Hough transform for ellipses. This was made practi-
The proposed ellipse-finding procedure is particularly cal by limiting the search space using a priori knowledge
suitable for the extraction of concentric ellipses with of the approximate ellipse parameters. The parameters
equal orientation but differing radii. A three-dimensional extracted by all three methods are tabulated in Fig. 12.
representation of the radius accumulator array for the Both the APR and ECR extensions of the ellipse-fintling
concentric ellipse set B is shown in Fig. 10. The major procedure successfully extracted all the ellipses in
PARTS. It can be observed that co-ordinate values
delivered by both versions of the three stage technique

.io
.+
.--
.- .-...-

Fig. 7 Subset of points labelled with centre of concenrric' ellip.se set A

Fig. 9 Subset of points labelled with centre ofsingle ellipse conrojdr

Fig. 10 Three-dimensional representation of radius uccumulutor urray


Fig. 8 Subset o/ points lubelled w i t h centre of concentric' ellipse s e t B for conc.entric ellipse set B

I E E PROCEEDINGS-E. Vol. 138. N O . 1. J 4 . V C 4 R Y 1991 33


differ at most by one pixel value from the values obtained finding procedure particularly when the image feature set
by the standard HT. In Fig. 11, the ellipses extracted by is cluttered with irrelevant data. In this case the nurriber
the APR extension are shown to fit the edge map repre- of feature points identified with a candidate ellipse ce.itre
sentation well. Spurious peaks were detected by each is likely to be much less than the total number of points
method, however they were rejected since the number of in the image feature set. Thus. processor time consumed
feature points labelled with the peak locations was low. by the orientation and radius-finding stages is low and is
roughly equally divided between each stage. On the 01 her
hand, when concentric ellipses exist in the feature set (as
in image PARTS), the level of point pairing carried out
during stages 2 and 3 is likely to increase and processor
time is more equally divided between all three stage:;. A
speed-up factor of 2201149 = 1.48 was obtained using the
APR extension over ECR. The relative saving obtained is
modest since the feature set was dominated by concentric
ellipses.
The counting of plaque on Petri dishes is a highly
labour intensive task in biological and medical work.
Ellipses were fitted to plaque contours in digitised images
of Petri dishes in an attempt to carry out a plaque coint.
Overlapping or connected plaque were characterised by
highly asymmetric ellipses. With this assumption in mind,
if the ratio of major to minor radii was R , then R plaque
were assumed to exist within the detected ellipse. A IT ag-
nified sector of the Petri dish, containing 84 plaque
(excluding plaque touching the image boundary) was
used as a test image, (Fig. 13). Inter-ellipse pairing during
centre-finding was reduced by setting d,,, to 8 pixels.
and d,,, and L were set to 1 and 6. respectively. The
ECR extension of the ellipse-finding procedure required
1740 seconds of CPU time to extract 81 plaque. Only 137
Fig. 11 Evtracted ellipses using A P R eztumioii oi-erluyed orrr c d y r seconds o f CPU time were required using the APR exien-
map sion to extract 82 plaque. These are shown overlayed on
the original test image in Fig. 14. A speed increasc of
The processor time required by the ECR and APR nearly 13 times by the APR extension over the ECR
extensions of the tristage technique to extract each ellipse extension highlights the computational savings that can
set are shown in the table of Fig. 12. Because the centre- be obtained in some practical situations using this
finding stage is applied only once by the APR extension, method of multiple shape extraction. However, shape
processor times required to extract the parameters of the extraction using accumulator peak removal may not be
concentric set B and ellipse 5 were less than that con- as robust. Reaccumulation of the parameter space after
sumed by the ECR extension. Note that the CPU time ellipse contour removal may impose an additional com-
shown for the APR extension for the extraction of con- putational load during centre-finding, but is likely to
centric set A includes the time required to apply the result in more accurate feature point labelling.
centre-finding transform. It is difficult to give clear guide-
lines as to the processor time consumed by each stage of
the tristage technique since this depends on the nature of L

the feature space and on the original image. Processor


time consumed is generally dominated by the centre-

~~ ~ ~~

Ellipse Method Centre Orientation Major Minor Points CPU


yo Yo e radws radius (secs)
1 SHT 142 195 26 41 15
ECR 143 196 26 41 15
APR 143 196 26 41 15
2 SHT 142 195 27 33 11 ~

ECR 143 196 26 32 11 338 102


APR 143 196 26 32 11 338 101
3 SHT 141 51 153 33 23
ECR 142 51 154 34 23
APR 142 51 154 34 23
4 SHT 142 51 153 38 26 - ~

ECR 142 51 154 38 27 362 85


APR 142 51 154 38 27 327 38
5 SHT 145 152 145 22 15 - -
ECR 145 152 146 22 15 122 23
APR 145 152 145 22 15 163 10

Fig. 12 Standard H T ( S H T ) , ellipse conrour remoral ( t C R ) und


accumulator peak remocal ( A P R )algorithms upplird to imuge P.4R T S
Total CPU time using ECR extension: 220 seconds (includes 10 seconds reaccumu-
lating the centre array after ellipse 5 was extracted) Total CPU times using APR
extension: 149 seconds. Ellipses 1 and 2 constitute concentric set A whereas
ellipses 3 and 4 constitute concentric set B. Fig. 13 Digitised segmerit o f P e r r i dish
ing peaks representing ellipse centres in the accumulator
array was computationally less demanding than dele1ing
the contour of the ellipse in feature space. However, the
latter extension is likely to be more robust. A practical
application of the technique to count plaque sites on a
Petri dish demonstrated that removing peaks in the accu-
mulator array achieved a speed-up factor of nearly 13
times over ellipse contour removal. The technique was
also demonstrated to successfully extract ellipses assxi-
ated with an arrangement of mechanical parts.
Problems that remain are the excessive computational
effort incurred during centre-finding owing to the irrele-
vant point-pairing that occurs between different ellipses
and other shapes. This is experienced, to a degree, in later
stages when concentric ellipses exist in the feature space.
This problem may, to an extent, be resolved if some form
of image segmentation is applied in conjunction with
edge detection. Future work will continue to investigate
these possibilies.

6 References
Fig. 14 Extracted plaque contours using E C R el-tension
I YANG, H.S.. and KAK, A.C.: ‘Determination of the identity, .,osi-
tion and orientation of the topmost object in a pile’. Proc. 01 3rd
5 Conclusions workshop on computer vision: Representation and control. Bellair,
Michigan, 1985, pp. 3 8 4 8
7 BROOKS. R.A.: ‘Symbolic reasoning among 3-D models and 2-D
A tristage Hough transform for ellipse extraction has
been presented. Candidate ellipse centres were found images’. Artificial Intelligence. 1981. 17, pp. 285-348
3 PAVLIDIS. T.: ‘Algorithms for shape analysis of contours and
using an improved centre-finding technique and the waveforms’, / E € € Trans., 1980, PAMI-2, (4), pp. 301-312
remaining parameters were extracted in two sequentially 4 ILLINGWORTH. J.. and KITTLER, J.: ‘A survey of the Hough
executed stages using one- and two-dimensional Hough transform‘. Computer Vision. Graphics and Image Processing, 988.
parameterisations. A qualitative comparison has shown 44,pp. 87-1 16
5 SKLANSKY, J . : ‘On the Hough technique for curve deteclion‘.
that the new technique appears to solve many of the IEEE Trans., 1978, C-27, (10). pp. 923-926
problems associated with previous ellipse-finding 6 TSUJI. S.. and MATSUMOTO, M . : ’Detection of ellipses by niodi-
schemes. Parameterisations based on the polynomial rep- fied Hough transformation’, I E E E Trans., 1978, C-27, (81, pp. 777-
resentation of an ellipse can require large accumulator 78 1
7 TSUKUNE, H.. and GOTO. K.: ‘Extracting elliptical figures :ram
arrays whose ranges are difficult to define. Computa-
an edge vector field‘. Proc. of IEEE Computer Vision and Partern
tional savings in the tristage technique are significant Recognition Conference, Washington, DC. 1983. pp. 138-141
since the dimensionality of the accumulator arrays used 8 NAGATA. T., TAMURA, H., and ISHIBASHI. K.: ‘Detection of an
in the tristage technique is restricted to two and the ellipse b> use of a recursive least squares estimator’, J . Robotic
range of parameter values is predefined. Schemes that Systems. 1985, 2, (2), pp. 163-177
9 YUEN, H.K., ILLINGWORTH, J.. and KITTLER, J.: ‘DeteLtinp
estimate ellipse orientation and radii using least squares partially occluded ellipses using the Hough transform’, Image and
procedures can suffer from the presence of outliers. These, Vision Computing, 1989, 7, ( l ) ,pp. 31-37
it was shown, do not seriously affect the robustness of the I O ILLINGWORTH. J., and KITTLER. J.: ‘The adaptive Hough
new technique. transform‘. I E E E Trans., 1987, PAMI-9, ( 5 ) ,pp. 69Ck697
Although the likelihood of a systematic propagation of 1 1 VAN VEEN. T.M.. and G R O E N , F.C.A.: ‘Discretization errors in
the Hough transform’, Pattern Recognition. 1981. 14, pp. 137-145.
errors increases with the number of stages required, 12 ROGERS, D., and ADAMS, A.: ‘Mathematical elements for com-
multistage parameter estimation can result in a greater puter graphics’ (McGraw-Hill, New York, NY, 1976)
degree of control and flexibility in shape extraction. The 13 GERIG, G., and KLEIN. F.: ‘Fast contour identification thrt,ugh
tristage technique was shown to be particularly suited to efficient Hough transform and simplified interpretation strategy’.
Proc. 8th Int. Conf. on Pattern Recognition. Paris, October 986.
extract concentric ellipses. pp. 498-500
Two extensions of the technique to multiple ellipse 14 CANNY, J.: ‘A computational approach to edge detection’. I E E E
extraction have been presented. I t was shown that delet- Trans.. 1986. PAMI-8, (6),pp. 679-698

35

Das könnte Ihnen auch gefallen