See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/5576362

Region Growing: A New Approach

Article in IEEE Transactions on Image Processing · February 1998

DOI: 10.1109/83.701170 · Source: PubMed

CITATIONS READS
561 2,992

2 authors, including:

S. A. Hojjat
Samsung
82 PUBLICATIONS 1,971 CITATIONS

SEE PROFILE

All content following this page was uploaded by S. A. Hojjat on 27 May 2014.

The user has requested enhancement of the downloaded file.

 Region Growing: A New Approach
S A Hojjatoleslami and J Kittler
Department of Electronic & Electrical Engineering,
University of Surrey, Guildford, UK, GU2 5XH

VSSP-TR-6/95

Abstract
Accurate segmentation of images is one of the most important objectives in
image analysis. The two conventional methods of image segmentation, region
based segmentation and boundary nding, often su er from a variety of limi-
tations. Many methods have been proposed to overcome the limitations but the
solutions tend to be problem speci c. Here we present a new region growing
method with the capability of nding the boundary of a relatively bright/dark
region in a textured background. The method relies on a measure of contrast
of the region which represents the variation of the region gray level as a func-
tion of its evolving boundary during the growing process. It helps to identify
the best external boundary of the region. The application of a reverse test
using a gradient measure then yields the highest gradient boundary for the
region being grown. A number of experiments have been performed both on
synthetic and real images to evaluate the new approach. The proposed scheme
can be categorized as a region based segmentation method which uses gradient
information to specify the boundary of a region. The main strength of the
method is its ability to segment out from a textured background a bright/dark
region with fuzzy boundaries as well as its simplicity and immunity to inten-
sity changes.

1
1 Introduction
The segmentation of regions is an important rst step for a variety of im-
age analysis and visualization tasks. There is a wide variety of image seg-
mentation techniques in the literature, some considered general purpose and
some designed for a speci c class of images. Segmentation techniques for
monochromatic images can be categorized into two di erent approaches [2].
One is region based, which relies on the homogeneity of spatially localized
features, whereas the other is based on boundary nding, using discontinu-
ity measures. The two methods exploit two di erent de nitions of a region
which should ideally yield identical results. Homogeneity is the characteristic
of a region and nonhomogeniety or discontinuity is the characteristic of the
boundary of a region.
If a region is homogeneous with relatively high contrast, the detection
of the region boundary becomes a simple task using any of the two conven-
tional methods. But the problem arises when the high frequency information
characteristic of a boundary is missing or is unreliable, with the consequence
that the region is not well de ned and an uncertain boundary exists. In such
situations, boundary nding methods fail, especially in the presence of noise.
Although, region based techniques are less a ected by noise, they commonly
su er from the problem of over-growing into neighbouring regions or back-
ground specially when these are textured. Furthermore, since conventional
boundary nding methods rely on changes in gray level, rather than on their
actual values, they are less sensitive to changes in image contrast than the re-
gion based segmentation methods. Also boundary nding methods in general
do a better job of boundary localization [2, 5].
Many studies investigating the properties of the two approaches for im-
age segmentation have been reported [1, 3, 4, 6, 7, 12, 14, 15]. As the two
methods use complementary information, they involve con icting objectives
and therefore their direct comparison is not straightforward. Most of the
reported techniques rely on a region growing method and use some discon-
tinuity measures as a stopping criterion to avoid the problem of merging
two neighbouring regions or over-growing into the background. The quality
of these techniques is highly dependent on the edge operator used [7, 8] as
a measure of discontinuity. Other approaches use the slope of a local pla-
nar approximation of the image surface. The idea is to test the hypothesis
that the slope of the plane has changed which would be characteristic of the

2
boundary between two neighbouring regions. Fitting a plane to image inten-
sities over a set of pixels requires information about the region which is not
always accessible in real situations. Consequently, the methods often exhibit
poor performance in de ning the boundary. A good survey of di erent ap-
proaches to region growing, their capabilities and limitations is presented by
Haralick [8].
We present here a new idea for region growing by pixel aggregation which
uses new similarity and discontinuity measures. A unique feature of the pro-
posed approach is that in each step at most one candidate pixel exhibits the
required properties to join the region. This makes the direction of the grow-
ing process more predictable which is the most important characteristic of
our method. The novel growing procedure o ers an ideal framework in which
any suitable measurement can be applied to de ne a required characteristic
of the segmented region. We use \contrast" and \gradient" as sequential
discontinuity measurements derived by the region growing process whose lo-
cally highest values identify the external boundary and the highest gradient
boundary of each region, respectively. The method rst nds the location of
the highest contrast boundary which is the external boundary of a region.
Then a reverse test using the \gradient" measure is applied to produce the
highest gradient region. Since the two measurements are based on gray level
di erence, the method is not sensitive to intensity changes. This contrasts
with the existing region growing techniques [8, 10, 14]. The method is very
e ective in de ning the boundary of a region with fuzzy edges located in a
textured background.
Like the existing procedures, the proposed method in this paper has not
a universal capability, but, on the other hand, it does appear to have a
fairly wide application potential, especially in medical image analysis, where
the areas corresponding to a tissue of interest appear as bright/dark objects
relative to the surrounding tissues and both the foreground and background
tissues exhibit textural variations.
The concept of the method is presented in the next two sections. The
similarity measure used by the method is presented in Section 2. Section 3
introduces the two di erent discontinuity measures, \contrast" and \gradi-
ent" and considers their behaviour on a Gaussian shape image. Section 4
considers the behaviour of the measurements on noisy or textured images
and illustrates that our method is independent of the choice of a starting
point. Section 5 demonstrates the capability of our method on a set of real

3
images.

2 Growing Process
The concept of our method, like other region growing methods by pixel ag-
gregation, is to start with a point that meets a detection criterion and to grow
the point in all directions to extend the region. The choice of the starting
point will be discussed later.
Let us assume that the process starts from an arbitrary pixel. The pixel
is labeled as a region which grows based on the similarity measure used. In
our approach, a boundary pixel is joined to the current region provided it
has the highest gray level among the neighbours of the region. This induces
a directional growing such that the pixels of high gray level will be absorbed
rst. Then the pixels with monotonically lower and lower gray levels will
join the region. When several pixels with the same gray level jointly become
the candidates to join the region, the rst-come rst-served strategy is used
to select one of them. This makes the region more compact, particularly in
situations where the gray levels of the background or the region pixels are
very homogeneous.
We generate gray level, gradient and contrast mappings during the grow-
ing process. The mappings are very similar to the mapping used in the mode
separating (MODESP) procedure proposed by Kittler [11] for cluster anal-
ysis. MODESP method is a clustering procedure based on the mapping of
data points from an N-dimensional feature space onto a sequence in which
each cluster in the space appears as a mode in the mapping. Separating
surfaces between the modes in the N-dimensional space are derived from
the points associated with distinct modes in the one-dimensional mapping
function. MODESP has never been used for the segmentation of spatially
indexed data and the only similarity of our method with MODESP is the
mapping used to monitor the growing process.
Consider Figure 1(a) which shows a small subimage with a single bright
blob. To present the concept of the growing process on this data, let us
assume that its starting point y1 is the pixel with the maximum gray level
of the subimage and de nes a nucleus of the blob region. The sequence of
pixels joining the region is y2; y3; y4 and so on. The graph of gray levels
associated with the sequence of candidate pixels for the region generated by
the growing process is shown in Figure 1(b). The mapping shows that the
4
 (a)

(b)

Figure 1: (a) Topographical surface of a microcalci cation in a homogeneous

background and (b) Mapping of gray levels of the region during the growing
process.

gray levels decreases from the highest value in the region to the background.
A similar mapping can be obtained for any measurement de ned on the
growing region. The mapping function de ned on the sequence of pixels
joining the growing region characterizes the variation of each measurement
in the spatial domain. Di erent criteria can be used to stop the growing
process and to apply a reverse check on the relevant measurements to detect
the region boundary. We use the maximum possible size N of a region to
stop the process. However, other criteria, such as minimum size of neigh-
bouring region or maximum di erence between the current candidate and
the maximum gray level inside the region can also be applied to stop the
growing process. We used the latter criterion for the segmentation of calci -
5
cations in mammographic images [9]. The size of a region is simply measured
by counting the number of pixels in the mapping. This can be formalized
by the following rule: consider the current pixel generated by the similarity
measure as a region candidate provided its index number i satis es:
i<N (1)
where N is the maximum expected size (number of pixels) of the region of
interest.
In the next section we consider the use of two measurements as charac-
teristic features of a region to nd its best boundary among all the candidate
boundaries considered during the growing process. The applied measure-
ments are not sensitive to the selected threshold N . Hence, the check against
the threshold is introduced only to avoid unnecessary growing into a neigh-
bouring region or homogeneous background.

3 Discontinuity Measures
For segmentation purposes we de ne region of interest as a gray level blob,
exhibiting a high contrast relative to its background. The best boundary for
a region is a connected boundary of the highest gradient. Thus we search
for the highest gradient boundary where the contrast is also high. Two
di erent measurements called \contrast" and \gradient" are used as criteria
to characterize the properties of the current region and its boundary during
the growing process.
In the description of the proposed algorithm, four di erent boundaries are
mentioned which are de ned as follows: Current boundary is the set of pixels
adjacent to the current region during the growing process. Internal boundary
is de ned as the boundary produced by the set of connected outermost pixels
of current region. The current region and the two boundaries are dynamically
changing during the growing process. Gradient boundary is the boundary of
the nal output region produced by the region growing method. External
boundary is the outermost boundary at which the growing process stops.
The contrast measure c(i) for a region containing i pixels is de ned as
the di erence between the average gray level of the region and the average

6
of its current boundary. This is expressed by:
X
c(i) = 1i y ? k ?
i
1 X
k

t =1
t
i = +1 y
t i
t (2)
where y1; y2; :::; y is the sequence of pixels forming the current region and
i

y +1; y +2; :::; y is the set of its current boundary pixels. This measurement
i i k

captures the local contrast of the region based on all the gray levels inside
the region and its boundary.
Let us recall that the algorithm always searches for the highest gray level
in the boundary. The pixel with the highest gray level is then added to the
growing region which systematically replaces the region boundary with pixels
of lower intensity values. The region growing will produce increasing contrast
measure values as long as the growing region continues subsuming high in-
tensity pixels of the bright blob. Once it starts growing into the background,
the rate of gray level decrease for the boundary will be less than that for its
region, and consequently the contrast will commence decreasing. Hence, the
maximum of this measurement during the growing process corresponds to
the point when the process starts to grow into the background. The result of
the segmentation based on the maximum contrast is the external boundary
of the region.
We approximate the gradient of a region using the di erence between gray
level average of the current internal boundary and the average of the current
boundary. The mapping of this measurement during the growing process
shows the pixel by pixel variation of the boundary gradient of the evolving
region. If the region is smooth the gradient is a reliable measurement to
de ne the point where the growing process should be stopped.
One could say that contrast is equal to the gradient measure biased by
the gray level average of its internal pixels. This becomes apparent by rear-
ranging Equation (2) as:
X
c(i) = g(i) + 1l y
l

t (3)
t=1
where g(i) is the mean of the internal-current boundary pixel pair gradients
and y1; y2; :::; y are the gray levels of the pixels inside the region excluding
l

the internal boundary pixels.

The measurement is a very sensitive approximation of the gradient, be-
cause only the di erence between two neighbouring boundaries is used to
7
de ne it. If the region is very noisy or textured the gradient could exhibit
false peaks. It is particularly a ected by noise or texture when the size of the
region is very small and therefore the number of pixels in the boundary is
low. Hence, we use the contrast measurement to nd the external boundary
and the last local maximum of the gradient to determine the highest gradient
region to terminate the growing process.
However, every local peak of the gradient can be used to segment out a
distinct region which meaningfully corresponds to the information conveyed
by the internal parts of the region. Furthermore, for a relatively homoge-
neous region, the global maximum of the gradient is uniquely de ned so that
without using the contrast measure, the region can be segmented.
Commonly the external boundary and the actual boundary are not very
far away from each other. The di erence is especially low when a bright
region has a vary sharp edge and it is very high for fuzzy edge regions. We
used a Gaussian shape image which has a very extensive external boundary
as compared to its gradient boundary, to show the e ect of the two measure-
ments. However in applications where strong edges exist the di erence is not
signi cant.
Theoretically the highest gradient of a Gaussian shape is located one
standard deviation from the mean. Equation (4) de nes a two dimensional
Gaussian shape:
? [ x?mx + y?my ]
1 ( )2 ( )2

g(x; y) = M exp 2
x 2
y 2
(4)
where m ,m specify the x; y location of the centre of the Gaussian blob
x y

and  ,  specify the spread of the gray level function. Constant M is

x y

used to normalize the output to the maximum gray level range. However
the e ect of quantization is that only an approximation of the Gaussian
image is obtained. For the sake of simplicity identical spread parameters are
used,  =  . Hence in polar coordinates the shape can be represented by
x y

Equation (5).
r?r
g(r) = M exp? [  ]
q q (5)
1 ( 0) 2
2

where r0 = (m2 + m2 ), r = (x2 + y2) and  =  =  . Maximum

x y x y

gradient for a Gaussian shape is a circle with radius , centred on r0 .

A Gaussian shape image with a variance of 25 pixels,  = 25, Figure 2(a),
is used to demonstrate the relationship of the two boundaries. Let the grow-
ing process starts at the highest gray level in the region, 255. The gray level,

8
contrast and gradient mappings during the growing process are shown in Fig-
ure 2(e). The gray level mapping shows that the gray levels of the sequence
of pixels joining the region from the starting point monotonically decrease to
zero corresponding to the background. As a result of the directional growing
process, the shape of the region for the Gaussian shape is circular even when
the process continues to absorb the zero gray levels in the background. This
is apparent by considering Figure 2(d) and noting that the gray level of all
the candidate pixels beyond pixel number 21772 is zero. As one might expect,
contrast commences from a minimum and smoothly increases to a maximum
at point 6685 and then decreases. The maximum contrast point corresponds
to a circular region with the radius of 46:13 pixels which is approximately
1:85 in the Gaussian image. The result of segmentation using this point is
shown in Figure 2(c). As can be seen, the segmented region based on the
contrast matches with the external boundary of the Gaussian shape at 1:85.
The mapping of gradient starts from low values increasing to a maximum
at pixel number 2000 and then decreasing again to zero. The maximum
gradient point corresponds to a circular region with the radius of 25:23 pixels
which is approximately 1:009 in the Gaussian image. This result agrees
well with the maximum gradient region of an analog Gaussian shape. The
slight di erence is caused by the e ect of quantization and the fact that
our method uses the di erence between the mean of two completely closed
contour boundaries to calculate
p the contrast, so the e ect of diagonal pixels
the distance of which is 2 is the same as that of pixels located in the
neighbouring position with distance 1. The result of segmentation using this
criterion is shown in gure 2(b).
We advocate the use of gradient as the nal result of segmentation and
apply the contrast measure to nd the external boundary of the region of
interest. Note that for sharp edge regions, the results of segmentation using
the contrast and gradient measures are similar. In Section 4, we consider
the e ect of noise and texture on the results produced by applying the two
complementary measurements.

4 Textured Image
In this section, the aim is to consider the performance of the procedure in
situations where a noisy textured region is located in a textured background.
A part of a road centre line shown in gure 3(a) is used as the input. As can
9
be seen, the dynamic gray level range of the region is quite high.
A point located at (60; 108) with the gray level of 175 is used as a starting
point to segment the region. The gray level mapping is shown in gure 3(e)-
top. Note that, the mapping exhibits uctuations during the growing process
two of which are particularly noticeable at pixel numbers 1100 ? 1400 and
7100 ? 7700. The artifact of each peak is a distinct valley in the contrast
and gradient mappings which occur in our example at pixel numbers 1300
and 7465, respectively, see Figure 3(e)-bottom. These two measures decrease
inside the area of locally high gray level and increase after it has been covered
(based on the gray level changes in the region and its boundary). The maxi-
mum contrast measure de nes the external boundary of the region containing
18240 pixels, see gure 3(b). Region's texture produces more uctuations in
the gradient measure than in the contrast measure. The gradient mapping
shows that the di erence between the global maxima which speci es the
boundary of the region and local maxima which are created by texture and
noise is not high enough to rely fully only on the gradient measure. In con-
trast, the global peak of the contrast measure is not a ected by the texture in
the region. Hence, as mentioned before, a reliable maximum gradient point is
the last gradient peak which is located before the global peak of the contrast
measure. This is located at pixel number 12031, see gure 3(e)-bottom. The
segmented region based on this point is shown in gure 3(c). This region is
in a full agreement with the result of human visual segmentation.
If the growing process terminates at any local gradient peak, a region
which is particularly acceptable by human visual system will be segmented.
Figure 3(d) shows a region segmented out based on the rst local maximum
of the gradient of value 37:44 containing 511 pixels. The region can be
characterized as a subregion of region shown in Figure 3(c).
Gray level, contrast and gradient mappings obtained from any starting
point inside the region will eventually start appending the same pixels once
all the brighter pixels are absorbed by the growing process. This happens
as a result of the strategy of always appending the pixels of the largest
gray level in the boundary to the current region. Hence, the segmentation
results are the same regardless of which starting point is used inside the
region. Figure 4 shows the mapping produced during the growing process
on the road centre line image using two di erent starting points located at
(362; 90) and (60; 108) with the gray levels of 171 and 175, respectively. The
two starting points are marked on the image, Figure 3(a). The e ect of

10
non-homogeneity of the region is noticeable in the mappings. Note that,
the mappings for di erent starting points are di erent while the growing
process absorbs brighter pixels inside the region. But, after these pixels have
been assigned, the mean of the gray levels inside the current region starts
decreasing with a lower rate than that of the boundary gray levels. Hence,
as expected, both the sequence and the corresponding measurements will
converge to the same point and the mapping will exhibit the same behaviour
thereafter. For the road centre line, the mapping functions become identical
after pixel number 1000, Figure 4. The highest contrast for the region is
always located at pixel number 18240 with the gray level of 77. The highest
gradient region containing 12031 pixels with the minimum gray level of 112
is speci ed by the gradient measure maximum of 41:72.
The use of the two contrast measures produces a unique region indepen-
dent of the starting point. This will be further clari ed in the next section
when we consider the region growing procedure on another real image.

5 Experiments on Real Images

This section shows the performance of our method on medical images where
each region can be categorized as a bright blob separated from its neighbour
by a low gray level boundary. First we show that our method is not sensitive
to threshold N of rule (1). We then give an interpretation of the gray level
mapping when N is very high in comparison to the size of the region of
interest. The segmentation result obtained for Magnetic Resonance Imaging
(MRI) and Computer Tomography (CT) Scan images of a head are presented
and discussed. For each region we specify an arbitrary internal starting point.
Figure 5 shows an MRI of head. The aim is to outline the corpus callo-
sum, brain stem and cerebellum. As we can see edges are very fuzzy at the
boundaries of these regions. Consequently, neither boundary nding methods
nor region based methods can reliably determine the boundaries.
We rst start with the segmentation of brain stem. A very high thresh-
old, N = 20000, in comparison to the size of the region is used to provide
an opportunity to consider the behaviour of the di erent discontinuity mea-
surements in relation to the neighbouring regions. The gray level, contrast
and gradient mappings obtained during the growing process are shown in
Figure 6. As can be seen, the highest contrast measure determines the loca-
tion of the external boundary and the last gradient measure maximum before
11
the maximum contrast point speci es the best boundary for the region. The
region segmented out using the method is shown in gure 5(b).
The gray level mapping shows local valleys which are induced by the gray
levels at the boundary of two neighbouring regions. Each visible valley in
the gray level mapping is the result of the switch between the absorption of
decreasing gray levels of the pixels in the boundary of the region being grown
and the absorption of pixels of increasing gray levels leading to the nearest
local peak of the neighbouring region. As mentioned before, the gradient of
the gray level mapping on the left side of the valley is related to the rate of
gray level decrease in the boundary of the region and the size of its current
boundary. The size of the region a ects the rate, because the number of
pixels in the boundary is a function of it. Hence, the bigger the size, the
lower the gradient. The gradient on the right side of the valley is related
to the rate of gray level increase along the pathway forged by the growing
process forwards a local hill of a neighbouring region. The latter is very sharp
because the growing process takes a single pixel wide path to the top of the
hill and then continues to cover its surface. Thus, the di erence between a
valley minimum and the following peak in the gray level mapping shows the
di erence between the maximum gray level of the hill and the maximum gray
level at which the two neighbouring regions meet. If the di erence is quite
high and the number of pixels in the new region is big enough, it is a strong
clue for the existence of a new signi cant region. Otherwise the new hill is a
local peak or noise in the region being grown.
The e ect of these variations is even more clear is the contrast mapping.
The growing into a neighbouring region causes a more rapid increase in the
mean of the gray levels of the pixels in the region's boundary than that of the
region itself. So, a local peak in the grey level mapping causes a local valley
in the contrast mapping and consequently a local peak. The local peak in the
contrast measure corresponds to the external boundary of its corresponding
region. Its related gradient peak will then specify the best boundary to the
region. The contrast mapping in Figure 6 shows the sequence of such peaks
and valleys.
Figure 5(c) shows the boundary corresponding to the last local gradient
maximum before the second distinct contrast peak which segments a region
containing 5677 pixels. The two gradient maxima located before the second
and third contrast peaks are located at point 11904 and 14448 (see Figure 6-
bottom). The corresponding boundaries are shown in Figure 5(d) and (e),

12
respectively. As mentioned before, each boundary has a meaningful informa-
tion regarding di erent possible regions produced by the process which can
be of interest in target detection.
Segmentation results of di erent parts of the MRI image are shown in
Figure 7(a). For each segmented region in the image a starting point is
selected. We tested many di erent starting points in di erent positions of
each region but the segmentation results were similar. The independence of
the segmentation results from the choice of a starting point is an important
characteristic of our approach. Figure 7(b) shows the result of segmentation
based on the contrast mapping. As can be seen, the external boundary of each
region is well characterized. Figure 7(c) and (d) show the distinct regions
produced by applying each of the two measurements separately.
All the considerations which are presented here are applicable to segment-
ing out a dark region if the whole process is reversed. In such a case, a region
with the minimum gradient is the result of the segmentation method. We
show this by applying the method to segment out the cavities in a CT-Scan
image. Figure 8(a) shows a CT-Scan image. The segmentation results of our
method using three arbitrary starting points, one in each cavity, are shown
in Figure 8(b). The results are again in full agreement with the results of
human visual segmentation.

6 Summary and Conclusions

We presented here a new idea for region growing by pixel aggregation, using
novel similarity and discontinuity measures. A unique feature of the proposed
approach is that in each step at most one candidate pixel will exhibit the
required properties to join the region. This makes the direction of the growing
process more predictable. Two new discontinuity measures named \contrast"
and \gradient" which use gray level di erence information to produce the
nal segmentation result are proposed and their properties analyzed.
Since the growing process is directional, i.e. pixels join the grown region
according to a ranking list and moreover the discontinuity measurements are
tested pixel by pixel, the method does not necessarily include all the pixels
with the same gray level to the region. This contrasts with thresholding
methods where all the pixels exceeding a certain threshold are included in the
segmented region [13]. From an extensive experimental testing, our method
appears to be more reliable and consistent than the thresholding methods.
13
 The result of our method does not be appear to be e ected by the presence
of a reasonable amount of noise. Hence, it can be used for segmenting raw
images without any need to apply a smoothing lter or perform preprocessing
to improve the signal to noise ratio. This property of the proposed method is
in sharp contrast to standard segmentation techniques which are commonly
disturbed by noise.
As \contrast" and \gradient", which determine the segmentation result,
are di erential measurements, the method is una ected by intensity changes
which is the characteristic of conventional region growing techniques.
The most signi cant features of the method are:
 Independence from the starting point location.
 Insensitivity to a reasonable amount of noise, and to region and/or
background being textured.
 Since the method uses two discontinuity measurements which delineate
a region, it does not need any statistical information concerning the
region.
We should mention that the method is applicable for the segmentation
of bright/dark regions in a dark/bright background without using any priori
knowledge about the region.

References
[1] R Adams and L Bischof, \Seeded region growing",IEEE Trans. PAMI,
Vol. 16, No. 6, pp 641-647, 1994.
[2] D H Ballard and C Brown, \Computer Vision", Springer Verlag, 1982.
[3] P J Best and R C Jain, \Segmentation through variable-order surface
tting", IEEE Trans. PAMI, Vol. 10, pp 167-192, 1988.
[4] S-Y Chen, W-C Lin and C-T Chen, \Split-and-merge image segmentation
based on localized feature analysis and statistical tests", CVGIP, Vol. 53,
no. 5, pp 457-475, 1991.
[5] L S Davis, \A survey of edge detection techniques", CGIP, Vol. 4, pp
248-270, 1975.
14
[6] R M Haralick and I Dinstein, \A spatial clustering procedure for multi-
image data", IEEE Trans. Circuits and Systems, CAS-22, pp 440-450,
1975.
[7] R M Haralick, \Edge and region analysis for digital data", CGIP, Vol.
12, pp 60-73, 1980.
[8] R M Haralick and L G Shapiro, \Survey: Image segmentation tech-
niques", CVGIP, Vol. 29, pp 100-132, 1985.
[9] S A Hojjatoleslami and J Kittler, \Automatic detection of calci cation
in mammograms", IEE Fifth Int. Conf. on Image Processing and its Ap-
plications, pp 139-143, 1995.
[10] H Jiang, Ji Toriwaki and H Suzuki, \Comparative performance evalu-
ation of segmentation methods based on region growing and division",
Systems and Computers in Japan, Vol. 24, No. 13, pp 28-42, 1993.
[11] J Kittler, \A locally sensitive method for cluster analysis", Pattern
Recognition, Vol. 8, pp 23-33, 1976.
[12] T Pavlidis and Y-T Liow, \Interrating region growing and edge detec-
tion", IEEE Trans. PAMI, Vol. 12, pp 225-233, 1990.
[13] P K Sahoo, S Soltani and A K C Wong, \A survey of thresholding
techniques", CVGIP, Vol. 41, pp 233-260, 1988.
[14] X Yu and J YlaJaaski, \A new algorithm for image segmentation based
on region growing and edge detection", Proc. IEEE International Sym-
posium on Circuits and Systems, Vol. 1, pp 516-519, 1991.
[15] M A Wani and B G Batchelor, \Edge-region-based segmentation of
range images", IEEE Trans. PAMI, Vol. 16, pp 314-319, 1994.

15
 (a) (b) (c) (d)

Gray level mapping

200 Considered region 2

Maximum contrast region
Maximum gradient region
Gray level

1
100

0
20000 22500 25000
0
0 5000 10000 15000 20000 25000
Contrast and Gradient mappings
80

60
Difference measures

Contrast
Gradient
40

20

0
0 5000 10000 15000 20000 25000
Number of pixels

(e)

Figure 2: The segmentation results of a Gaussian shape image with  =

25. The region size criterion used is N = 25000. (a) Original image. (b)
Segmentation result based on the gradient measure. (c) External boundary
segmented by the maximum contrast point. (d) The boundary produced by
the region containing 25000 pixels. (e) Gray level, gradient and contrast
mappings obtained during the growing 16 process.
 (a) (b)

(c) Gray level mapping (d)

180

Considered region
140 Maximum contrast region
Gray level

100

60
0 5000 10000 15000 20000

Contrast and Gradient mappings

55
Last gradient peak
First gradient peak Global contrast peak
Difference measures

40

Contrast
25
Gradient
7465
1300

10
0 5000 10000 15000 20000

17
Number of pixels

(e)

Figure 3: The segmentation results of a part of road centre line. (a) In-
put image. (b), (c) Boundary produced by applying contrast and gradient
measures, respectively. (d) Segmentation results based on the rst local gra-
dient maximum at pixel number 511. (e) Gray level, contrast and gradient
mappings during the growing process.
 Gray level mapping
180

Starting point (60,108)

Starting point (362,90)

140
Gray level

100

60
0 5000 10000 15000 20000

Contrast mapping

45

Starting point (60,108)

Contrast

Starting point (362,90)

25

5
0 5000 10000 15000 20000

Gradient mapping
45

35
Gradient measure

25

Starting point (60,108)

Starting point (362,90)
15

5
0 5000 10000 15000 20000
Number of pixels

Figure 4: Mapping of gray level, contrast and gradient measures during the
18 at (362; 90) and (60; 108).
growing process for two starting points
 (a) (b)

(c) (d) (e)

Figure 5: (a) Original MRI image. (b) Segmentation result of brain stem. (c),
(d) and (e) Segmentation results based on di erent locally highest gradient
regions at pixel numbers 5677, 11904 and 14448, respectively.

19
 Gray level mapping

125

110
Gray level

95
Considered region
Maximum contrast region

80
0 4000 8000 12000 16000 20000
Contrast and Gradient mappings
30
Difference measures

20

10 Contrast
Gradient

0
0 4000 8000 12000 16000 20000
Number of pixels

(a)

Figure 6: The mappings for brain stem during the growing process starting
at pixel (304; 165), N = 20000.

20
 (a) (b)

(c) (d)

Figure 7: (a) segmentation results based on the highest contrast measure-

ment. (b) The external boundary of each region speci ed by the contrast
measurement. (c) and (d) Labeled regions based on the two measurements.
21
 (a) (b)

Figure 8: The segmentation results of a part of CT-Scan Image.

22

View publication stats