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

net/publication/258651893

Watershed Segmentation based on Distance Transform and Edge Detection

Techniques

Article  in  International Journal of Computer Applications · August 2012

DOI: 10.5120/8259-1792

CITATIONS READS

14 2,512

2 authors, including:

Dibyendu Ghoshal
National Institute of Technology, Agartala
25 PUBLICATIONS   69 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Conceptual Design of a Ramp Type Escalator Utilizing Magnetic Levitation Concept View project

All content following this page was uploaded by Dibyendu Ghoshal on 27 July 2016.

The user has requested enhancement of the downloaded file.

 International Journal of Computer Applications (0975 – 8887)
Volume 52– No.13, August 2012

Watershed Segmentation based on Distance Transform

and Edge Detection Techniques
Pinaki Pratim Acharjya Dibyendu Ghoshal
Assistant Professor, Department of CSE Associate Professor, Department of ECE
Bengal Institute of Technology and Management National Institute of Technology
Santiniketan, West Bengal, Pin - 731236 Agartala, Tripura, Pin - 799001

ABSTRACT transmission of them through space media. The segmentation

An edge detection algorithm for digital images is proposed in of low contrast images have always posed problem due to
this paper. Edge detection is one of the important and most processing problem arising from inability to directly apply the
difficult tasks in image processing and analysis. In images watershed algorithm on the image. Rather it has been found to
edges can create major variation in the picture quality where be more suitable to apply watershed to the gradient of the
edges are areas with strong intensity contrasts. Edges in image instead of the original image [20]. In this approach, the
digital images are areas with strong intensity contrasts and a original minima of catchment basins correlate with the small
jump in intensity from one pixel to the next can create major values of the image gradient value. Thus the study of the
variation in the picture quality. This paper proposed an effect of application of varies modified version or
effective edge detection algorithm based morphological edge combination of morphological based watershed algorithm [9],
detectors and watershed segmentation algorithm using [14-19]on low contrast images of stellar objects have become
distance transform. The result confirms that the proposed worthwhile and interesting.
algorithm is found to yield satisfactory and efficient A good number of methods have been proposed and applied
segmentation of the digital images for edge detection. to solve the problem related to digital image segmentation
Experimental result presented in this paper is obtained by
[11], [13]. These techniques can be broadly classified into the
using MATLAB.
following categories as histogram based techniques, edge
Keywords based techniques, region based techniques, markov random
Edge detection, Segmentation, Distance Transform, field based techniques and hybrid techniques. All the above
Watersheds. techniques have got certain advantages as well as some
complexity compared to the others, the present technique
1. INTRODUCTION
Image segmentation has been an essential step in digital would fall within the group of processing having adaptive
image processed for most subsequent image analysis and histogram equalization as a preprocessing step for low
image understanding activities. In specific, many of the contrast stellar images.
prevailing method for image description and recognition [1-
2], image visualization [3-4], and object based image 2. THE WATERSHED SEGMENTATION
compression [5-7], highly depend on the segmentation result Watershed algorithm is based on morphological process
obtained from previous stage. The segmentation problem [10], although it can be mixed up with edge based segmentation to
[12] involves separating or partitioning of an image into a yield a hybrid technique. In geography, watersheds are the
number of homogenous and spatially connected group of ridge line that divides different areas called catchment basins
pixels called super pixels, so it has addition of two drained by different river systems. In mathematical
neighborhood segments produces a heterogeneous segment. morphology, a gradient image may be considered as a
As an alternative approach, segmentation can be thought of as topological surface where the numerical value of each pixel
a pixel labeling cum mathematical morphological process in indicates the evaluation of their points. The set of pixels along
the same that all the pixels belonging to the same which the gray levels changes sharply gives rise to an edge.
homogeneous region are assigned the same label and theme The watershed algorithm applies these ideas to solve a variety
and several techniques to define homogeneity of a region of image segmentation problems.
based on a particular objective of image segmentation
process.
Stellar images of various planets, stars, galaxies, comets,
nebula and other various extra terrestrial moving objects have
occupied a significant place in radio astronomy, space
research and preparation of distance space vehicle movements
(manual or robotized) [8]. The images of stellar objects have
been acquired either through satellite imagery or Hubble
space telescope. These images are very much prone to be
affected by a verity of noise like Gaussian noise, Rayleigh
noise, Impulse noise and Speckle noise. At the same time, the
contrast of these images have been generally found to be low Fig 1: Watershed segmentation-local minima of gray level
to very long distance and atmosphere turbulence and yield catchment basins, local maxima define the watershed
disturbance and variation of temperature during the lines.

6
 International Journal of Computer Applications (0975 – 8887)
Volume 52– No.13, August 2012

Assume, Mi where i= 1 to n be the set of coordinates points in

the regional minima (catchment basins), of the image P(x,y)
and C(Mi) be the coordinates points of catchment basins
associated with the regional minima Mi

(1)
Where,
T[n] = set of points in P(x,y) which are lying below the plane
p(x,y) = n
min, max = minimum or maximum gray level value.
n = stage of flooding varies from min + 1 to max + 1
Let Cn(M1) be the set of points in the catchment basin
associated with M1 that are flooded at stage n.

(2)
Where,

{
(3) Fig 2: (a) shows a binary image matrix, and (b) shows the
corresponding distance transform.
C[n] is the union of flooded catchment basin portions at the
stage n. 4. PROPOSED ALGORITHM
The image (Fig 3), being a distance object has low contrast.
Where, These may be various interferential factors in such type of
images such as non uniform illumination, variation in
(4) temperature of the ambience, atmospheric disturbance and
turbulence, variation in the intensity and contrast etc. All
these factors would enhance the difficulty of segmentation of
(5) galactic images. Thus a meaningful and effective measure is
warranted to alleviate the adverse influence. So some image
If the algorithm keeps on increasing flooding level then
processing method would have to be applied and the image
Cn(Mi) and T[n] will either remain constant or increase.
preprocessing stage comprises a number of methods such as
Algorithm initializes , and then
gray level transformation, contrast enhancement, image de
precedes recursively by assuming that at step n C [n - 1] has
noising, image smoothing, sharpening and so on.
been constructed.
This approach deals with catchment basins and watershed
Let, G is a set of connected components in T[n] and for each
ridge lines in an image by assuming it as a surface where light
connected component g G[n], there possibilities will arise.
pixels are low. But most important contrast of an image is
1. g ∩ C[n - 1] is empty. very important for image segmentation. It is the difference in
visual properties that makes difference between one object of
2. g ∩ C[n - 1] contains one connected component of C[n - an image from the background or we can say it is the
1]. difference in the color and brightness of the object. So for
3. g ∩ C[n - 1] contains more than one connected component satisfactory image segmentation image with satisfactory
of C[n - 1]. resolution is essential. In this approach, for low contrast
stellar images we have used watershed with markers. In the
3. WATERSHED SEGMENTATION initial step shown in figure 4, one stellar image is chosen,
where contrast of that image is low and appropriate area is
USING THE DISTANCE TRANSFORM cropped which is shown in figure 5. We need to adjust the
The distance transforms of a binary image is the distance from
contrast. Contrast adjustment is necessary for good image
every pixel of the object component which is black pixels to
understanding and analyzing. In figure 6, the image contrast is
the nearest white pixel. In binary images there are only two
adjusted with adaptive histogram equalization method. In
gray levels 0 and 1 where 0 stand for black and 1 stands for.
figure 7 we have applied threshold. In figure 8, we have clean
Only one catchment basin will appear in the topographic of a
up and overlay the perimeter on the original image by using
binary image surface only when two black blobs are
imoverlay function, This function takes input image and a
connected together. In this article,, the distance transforms
binary mask and generates an output image, where masked
along with canny edge detector have been used to preprocess
pixels have been replaced by a specified color. In figure 9, we
the image to make it suitable for watershed segmentation. In
apply markers for object and background. The extended
below, figure 2(a) shows a binary image matrix, and in figure
maxima operator is used to identify groups of pixels that are
2(b) shows the corresponding distance transform.
significantly higher than their immediate surroundings. Again
cleaning up and overlying is there in figure 10 and compute
the complement of the image so that the peaks become valleys
and modifies the image using imimposemin function. Finally
we will sum up the watershed transform in figure 11 along

7
 International Journal of Computer Applications (0975 – 8887)
Volume 52– No.13, August 2012

with consequence examination. In this paper we study this

approach with a low contrast image of a galaxy and detecting
it. The flowchart of the proposed method is given below in
figure 4.

Fig 4: Original image of Lena.

Fig 3: Flowchart of the proposed method.

5. RESULTS AND DISCUSSION

The experiments are carried out to evaluate the performance Fig 5: Traditional watershed segmented image using
of proposed method with existing watershed segmentation distance transform.
using distance transform method and with the usually used
differential edge detection operators such as Sobel edge
detector or LoG edge detector with the combination of
morphological watershed segmentation using distance
transform.
The proposed method has been tested on the image of Lena
with the size of {512 x 512} and with 256 gray levels (Figure
4). The final segmentation results are shown in figures (5, 6,
7, and 8). It can be observed from the final segmented images
that the final resultant image with proposed approach (Figure
8) produced better results compared to to traditional
watershed segmentation using distance transform method
(Figure 5) and the usually used differential edge detection
operators such as Sobel edge detector (Figure 6) or LoG edge
detector (Figure 7) with the combination of morphological
watershed segmentation using distance transform. Fig 6: Watershed segmented image using distance
According to results canny edge operator is more efficient for transform with sobel edge detector.
edge detection with watershed segmentation using distance
transform method where the other methods produced week
edges for Lena image. The histograms of the final resultant
images have been shown in figure (9, 10, 11, and 12). The
performance of Segmentation algorithm is measured with the
help of Entropy and (Table 1) as in term of visual quality of
the original image and the resulted image.

Fig 7: Watershed segmented image using distance

transform with LoG edge detector.

8
 International Journal of Computer Applications (0975 – 8887)
Volume 52– No.13, August 2012

Fig 8: Watershed segmented image using distance

transform with canny edge detector.
Fig 11: Histogram of watershed segmented image using
distance transform with LoG edge detector.

Fig 9: Histogram of watershed segmented image using

distance transform.

Fig 12: Histogram of watershed segmented image using

distance transform with canny edge detector.
Table I- statistical measurement
IMAGE ENTROPY ELAPSED
TIME
Watershed segmentation 4.3318 1.523536
using distance transform. seconds.
Watershed segmentation 4.6451 1.887422
using distance transform seconds.
with sobel edge detector.
Watershed segmentation 4.8717 1.776297
using distance transform seconds.
with LoG edge detector.
Fig 10: Histogram of watershed segmented image using Watershed segmentation 4.9610 2.002153
distance transform with LoG edge detector. using distance transform seconds.
with cannyl edge detector.

9
 International Journal of Computer Applications (0975 – 8887)
Volume 52– No.13, August 2012

6. CONCLUTSONS region based integration,” IEEE transections on

In this paper, an effective morphological edge detection Communications, Vol.43, 1995.
algorithm is proposed to detect image edge. The technique is [8] K. Haris, “A hybrid algorithm for the segmentation of 2D
very useful for Image segmentation and classification. The and 3D images,” Master’s thesis, University of Crete,
given experimental result shows that the algorithm produces 1994.
more efficient results compared to traditional watershed
segmentation using distance transform algorithm and the [9] R. Harlick and L. Shapiro, “Image segmentation
usually used differential edge detection operators such as technique,” CVGIP, Vol.29, pp. 100-137, 1985.
Sobel edge detector or LoG edge detector with the [10] Vicent L. Solille P, Watershed in digital spaces, “An
combination of morphological watershed segmentation using efficient algorithm based immersion simulations”, IEEE
distance transform. The detected edges are more sharp, Transections PAMI, pp. 538-598, 1991.
pinpointed and clear with abundant edge information.
Moreover, the proposed algorithm can filer the noise more [11] Gonzalez & Woods, Digital Image Processing, 3rd
effectively than traditional watershed segmentation using edition, Prentice Hall India, 2008.
distance transform algorithm.
[12] K. Haris,“Hybrid image segmentation using watersheds
7. ACKNOWLEDGEMENT and fast region merging,” IEEE Trans Image Processing,
The authors acknowledge the constant inspiration and 7(12), pp. 1684-1699, 1998.
encouragement from Professor P. K. Bose, Director, National [13] Jos B.T.M. Roerdink and Arnold Meijster.: “The
Institute of Technology, Agartala. watershed transform: Definitions, algorithms and
parallelization strategies,” Fundamental Informatics, Vol.
8. DEDICATION 41, pp.187-228, 2001.
One of the others (Dibeyendu Ghoshal) dedicates the entire
study to the loveliest and loving memory of his only one and [14] Hua LI et al., “An improved image segmentation
younger sister Kumari Sumita Ghoshal who herself was a gem approach based on level set and mathematical
of the scholars, a symbol of wisdom and art, peerless beauty morphology,” GREYC-ISMRA, CNRS 6072, 6 Bd
and simplicity, unfathomable knowledge and generosity. Maréchal Juin, 14050 Caen, France.

9. REFERENCES [15] Mahua Bhattacharya, Arpita Das, “A Study on

[1] P. Suetnes, P.Fua and A. J. Hanson, “Computational Seeded Region Based Improved Watershed
Transformation for Brain Tumor segmentation,” Indian
strategies for object recognition,” ACM Computing
Institute of Information Technology & Management,
Surveys, Vol. 24, pp. 05-61, 1992.
Gwalior Morena Link Road, Gwalior-474010.
[2] R. Besl, R. Jain, “Three dimensional object recognition,”
ACM Computing Surveys, Vol. 17, pp. 75-145, 1985. [16] D. Marr and E. Hildreth, “A theory of edge detection,”
Proc. R. Soc. London, no 207, pp. 187-217.
[3] K. Hohne, H. Fuchs, S. Pizer, “3D imaging in medicine:
[17] Chen Pan, Congxun Zheng, Hao-Jun Wang “Robust
Algorithms, systems, Applications”, Berlin, Germany,
Springer –Verlag, 1990. Color Image Segmentation Based On Mean Shift And
Marker-controlled Watershed Algorithm”, Second
[4] M. Bomans, K. Hohne, U. Tiede and M. Riemer, “3D International Conference on Machine Learning and
segmentation of MR images of the head for 3D display,” Cybernetics, Wan, ,pp 2752-2756, 2003.
IEEE Transactions on Medical imaging, Vol.9, pp. 253-
[18] F. Meyer, S. Beucher, “Morphological Segmentation”,
277, 1990.
Journal of Visual Communication and Image
[5] M. Kunt, M. Bernard and R. Leonardi, “Recent results in Representation, 1, pp. 21-46, 1990.
high compression image coding,” IEEE Trans. on
[19] N. Pal and S. pal, “A review on image segmentation
Circuits and Systems, Vol.34, pp.1306-1336, 1987.
techniques,” Pattern Recognition, Vol.26, pp.1277-1294,
[6] P. Willemin, T. Reed and M. Kunt, “Image sequence 1994.
coding by split and merge,” IEEE Trans. on Circuits and
Systems, Vol.34, pp.1306-1306, 1987. [20] K. Haris, G. Tziritas and S. Orphanoudakis, “Smoothing
2D or 3D images using local classification,” in Proc.
[7] F. D. Natale, G. Desoli, D. Glusto and G. Vernazza, EUSIPCO, Edinburg, U.K., 19994.
“Polynomial approximation and vector quantization: A

10

View publication stats