An efficient image segmentation approach through enhanced watershed algorithm

Computer Engineering and Intelligent Systems www.iiste.org
ISSN 2222-1719 (Paper) ISSN 2222-2863 (Online)
Vol.4, No.6, 2013 - Selected from International Conference on Recent Trends in Applied Sciences with Engineering Applications
1
An Efficient Image Segmentation Approach through Enhanced
Watershed Algorithm
Farheen K. Siddiqui Prof. Vineet Richhariya
M.Tech, Scholar, Computer Science Head, Computer Science
LNCT, Bhopal, India LNCT, Bhopal, India
Abstract - Image segmentation is a significant task for image analysis which is at the middle layer of image
engineering. The purpose of segmentation is to decompose the image into parts that are meaningful with respect
to a particular application. The proposed system is to boost the morphological watershed method for degraded
images. Proposed algorithm is based on merging morphological watershed result with enhanced edge detection
result obtain on pre processing of degraded images. As a post processing step, to each of the segmented regions
obtained, color histogram algorithm is applied, enhancing the overall performance of the watershed algorithm.
Keywords – Segmentation, watershed, color histogram
I. INTRODUCTION
Image segmentation is vital step in the process of image processing. Image segmentation is to divide an image into
number of regions so that each region gives information about an object or area of interest [1]. It partitions an
image into a set of non overlapping regions whose union is the entire image. The purpose of segmentation is to
decompose the image into parts that are meaningful with respect to a particular application, very much like the idea
of separating figure from ground.
The image properties consist of:
- Intensity values from original images, or
computed values based on an image operator
- Textures or patterns, unique to each type of region
- Spectral profiles that provide multi-dimensional image data
Elaborate systems may use a combination of these properties to segment images, while simpler systems may be
restricted to a minimal set on properties depending of the type of data available.
Most segmentation techniques are either region-based or edge based.
• Region-based techniques rely on common patterns in intensity values within a cluster of neighboring pixels.
The cluster is referred to as the region, and the goal of the segmentation algorithm is to group regions according
to their anatomical or functional roles.
•Edge-based techniques rely on discontinuities in image values between distinct regions, and the goal of the
segmentation algorithm is to accurately demarcate the boundary separating these regions.
II. LITERATURE REVIEW
The watershed algorithm is one of the most powerful morphological tools for image segmentation.
Hao Wei et al [10] presented an improved watershed algorithm for medical image segmentation. Firstly, an
iterative data-adaptive Gaussian smoother is used to smooth large scale details while suppress noises. Secondly,
the contour information of the original image is enhanced and revised by fusing the gradient with the detected
edges. Finally, valley-filling technique is used to control the number of the resultant watershed regions.
A new image segmentation based on reconstruction labeling watershed algorithm in color space [11] is proposed
which converts the RGB color image into a new color space image, and calculate gradients. Object regions are
extracted to compose a binary marked image through the use of morphological opening and closing
reconstruction, and gradient image is introduced for substituting the marked image. Finally, the watershed
transformation is carried out in the modified gradient image. This not only can overcome over-segmentation but
also retains the edge information of the object as much as possible.
Zhonglin et al [12] worked to obtain the size of contamination area of the insulator. They segmented the insulator
image with water-shed algorithm, and then the region growing algorithm was applied to process the segmented
image, which can guarantee the segmentation effectiveness and reduce the number of segmented regions so as to
enhance the segmentation results of the insulator image.
Jie Chen et al [13] discuss an image reconstruction and segmentation in an improved watershed algorithm by

2
using a plug-in function in flooding process. This extends the watershed transform for segmentation by allowing
the integration of a priori information about image objects into flooding simulation processes. Their method can
customize the initial seeding, region growing and stopping rules of the watershed flooding process according to
user-defined functions incorporating prior information.
Their method shows very low error rates compared with the recent literature and in comparisons with other
algorithms, while being easily adapted to new imaging contexts.
Shang et al [14] adopts a multi-resolution analysis theory, using wavelet trans-formation theory establishment
multi-resolution analysis framework on image segmentation. Multi-wavelet transform has some unique
advantages and good performance on edge detection and texture analysis. Their research uses multi-wavelet
transform to improve the over-segmentation
watershed transformation, to get meaningful image segmentation result.
III.WATERSHED TRANSFORMATION
The idea of watershed transform is straightforward by the intuition from geography. The main goal of watershed
segmentation algorithm is to find the “watershed lines” in an image in order to separate the distinct regions [9].
To imagine the pixel values of an image is a 3D topographic chart, where x and y denote the coordinate of plane,
and z denotes the pixel value. The algorithm starts to pour water in the topographic chart from the lowest basin
to the highest peak. In the process, we may detect some peaks disjoined the catchment basins, called as “dam”.
The watershed algorithm is one of the most powerful morphological tools for image segmentation.
Figure 1: Catchment basin and Watershed line
The watersheds may be obtained in two ways, either the boundaries of segmented regions, or the complement of
segmented regions. Beucher [2] categorized watershed algorithms into two groups. The algorithms in the first
group like immersion algorithm [3] simulate the flooding process. The immersion algorithm is one of the most
famous watershed segmentation algorithms. It offers an efficient way to extract watershed lines by simulating the
immersion process on gradient images. The second group of watershed algorithms aims at direct detection of
watershed lines.
Unlike the typical morphological filters, the watersheds transformation is not composed of the primitive
morphological operations. The initial concept of the watersheds transformation as a morphological tool was
introduced by H. Digabel and C. Lantuéjoul in [4]. Later, a joint work of C. Lantuéjoul and S. Beucher led to
the ‘inversion” of this original algorithm in order to extend it to the more general framework of grayscale images.
Later, watersheds were studied by many other researchers and used in numerous grayscale segmentation problems.
In this proposal, the efficient algorithm for watersheds suggested by Luc Vincent and Pierre Soille [5] is
reviewed briefly and used throughout the entire simulation.
Embedded Problems in Watershed Segmentation
The watersheds transformation is an effective method for extracting out continuous boundaries of each region and
gives solid results. However, applying it to the original image can cause undesired results.
(i) Oversegmentation
The watersheds transformation makes a number of regions as an output. For eg, a human can clearly see a woman
with background in Figure 2. This is because humans are capable of understanding the ‘semantics’ of a given scene;
however, this image has 2976 different regions after the watersheds transformation.

3
Figure 2: An Example of the Watersheds transformation on Gray-scale Image
This oversegmentaion problem comes mostly from the noise and quantization error. To eliminate the effect of
local minima from noise or quantization error on the final results, first, the gradient of the original image is
computed as a pre-processing and then the watersheds transformation is applied on the gradient of image.
Another approach is to apply a post-processing where a large number of regions are merged until the output
meets a given criteria which can be the number of regions or a dissimilarity value between homogeneous
regions.
(ii) Ambiguous Boundary on Homogeneous Regions.
Another reason why the watersheds transformation is applied to the gradient image is because the watersheds
transformation makes ambiguous boundaries occasionally. Figure 3 shows a simple case with a vague boundary.
As interpreted, it consists of 3 different homogeneous regions. If the watersheds transformation is applied directly
to the original image, then the output is just 2 different regions. The only marked boundary is made across the
middle homogeneous regions. It does not match the original boundaries.
Figure 3: Simple Illustration of Ambiguous Boundary Extraction:
(a)original, b) gradient, c) WS on original, d) WS on gradient
However, if the transformation is applied to the gradient of the original image, it can extract out proper
boundaries and recognize 3 different homogeneous regions.
IV. INTRODUCING THE CONCEPT OF COLOR HISTOGRAM
The JND color model in RGB space based on limitations of human vision perception as proposed in [6] is
briefed here for ready reference. The human retina contains two types of light sensors namely; rods and cones,
responsible for monochrome i.e. gray vision and color vision respectively. The three types of cones viz, Red,
Green and Blue respond to specific ranges of wavelengths corresponding to the three basic colors Red, Green
and Blue. The concentration of these color receptors is maximum at the center of the retina and it goes on
reducing along radius.
According to the three color theory of Thomas Young, all other colors are perceived as
linear combinations of these basic colors. According to [7] a normal human eye can perceive at the most 17,000
colors at maximum intensity without saturating the human eye. In other words, if the huge color space is
sampled in only 17,000 colors, a performance matching close to human vision at normal
illumination may be obtained. A human eye can discriminate between two colors if they are at least one ‘just
noticeable difference (JND)’ away from each other. The term ‘JND’ has been qualitatively used as a color
difference unit [8].
(a) Computing the Basic JND Histogram
Using this sampling notion and the concept of ‘just noticeable difference’ the complete
RGB space is mapped on to a new color space J r J g J b where J r , Jg and J b are three orthogonal axes
which represent the Just Noticeable Differences on the respective R,G and B axes. The values of J on each of the

4
color axes vary in the range (0,24) ,(0,26) or (0,28) respectively for red, blue and green colors. This new space is
a perceptually uniform space and offers the advantages of the uniform spaces in image analysis.
Histogram of an image manifests an important global statistics of digital images, which
can be used for a number of analysis and processing algorithms. In color image histograms, a large number of
colors may be present as required for representing real life images. All of these colors may not even be noticed as
different colors by normal human eye [8], hence as the first step the histogram on each of its axis has
been sampled suitably to accommodate all the human distinguishable colors.
(b)Histogram Agglomeration and Segmentation
Agglomeration in chemical processes attributes to formation of bigger lumps from smaller particles. In the
digital image segmentation, the similar pixels (in some sense) are clustered together under some similarity
criteria. And thus it was inspired that the agglomeration may contribute considerably in the process of color
image segmentation. The agglomeration techniques can be thought of as the powerful alternatives to the other
image thresholding techniques. After the compressed histogram of a real life image is obtained using the basic
JND histogram algorithm, the agglomeration technique can further be used to reduce the number of colors by
combining the smaller segments with similar colored larger segments. This helps in minimizing over
segmentation.
V. PROPOSED WORK
Proposed algorithm is based on merging morphological watershed result with enhanced edge detection result
obtain on pre processing of degraded images. Here preprocessing means image restoration. Hence it enhances
the watershed result. Color Histogram algorithm is applied to each of the segment as a post processing step
which further enhances the overall result.
The proposed segmentation algorithm is schematically illustrated in the block diagram shown below.
Figure 4: Flow chart of proposed work
Description of each block
(a) Input image (Degradation model)
In degradation model, the image is blurred using filters and additive noise via Gaussian Filter and Gaussian
Noise. Gaussian Filter represents the PSF which is a blurring function. The degraded image can be described by
the following equation
g = H*f + n
g is degraded/blurred image, H is space invariant function (i.e.) blurring function, f is an original image, and n is

5
additive noise.
Image deblurring can be done by the convoluting the image with 2-D Gaussian function.
(b) Canny edge detection and ringing effect
The Discrete Fourier Transform used by the deblurring function creates high frequency drop-off at the edges of
images. This high frequency drop-off can create an effect called boundary related ringing in deblurred images.
For avoiding this ringing effect at the edge of image, we have to detect the edge of an image. The edge can be
detected effectively using Canny Edge Detection method. It uses two different thresholds foe detecting both
strong and weak edges.
(c) Preprocessing
In preprocessing block we do image restoration i.e recovering the original image from the degraded image. Blind
Deconvolu-tion algorithm is applied to the blurred image. It is possible to renovate the original image without
having specific knowledge of degradation filter, additive noise and PSF.
(d) Morphological Reconstruction
The preprocessed image is morphologically reconstructed for smoothening the interior of the objects and to
preserve the boundary of the objects. The objects to be smoothened are subjected to a group of morphological
operators.
The basic morphological operators involved in this phase are dilation and erosion.
(e) Gradient Image Generation
The local gray-level variation in the image can very well be given by the morphological gradient. A gradient
helps detecting ramp edges and avoids thickening and merging of edges providing edge-enhancements. The
gradient image, G(f) is morphologically obtained by subtracting the eroded image, ε (f) from its dilated version,
∂(f) .
(f) Marker Extraction
Markers can be used to solve the over-segmentation problem whose goal is to detect the presence of
homogeneous regions from the image by a set of morphological simplifications. The Markers are connected
components belonging to an image, internal markers are inside each of the objects of interest and external
markers are contained within the back-ground. The resulting marker image M(f) is a binary image such that a
pixel is a marker (to be black) if it belongs to a homogeneous region, a pixel will be white if it does not belongs
to homogeneous regions.
(g) Image Segmentation
The Final Gradient image which is marker extracted is subjected to Watershed Segmentation. Watershed
segmentation produces a more stable segmentation of objects including continuous segmentation boundaries by a
concept of producing catchment basin and watershed line. Watershed and watershed lines are those points in a
topographic surface where a water drop placed would definitely fall into a single minimum and are equally likely
to fall in one or more of minimal points respectively.
(h) Computing Color Histogram
To each of the segmented regions obtained, color histogram algorithm is applied i.e computing the JND
histogram followed by the agglomeration of the histograms which can be thought of as the powerful alternatives
to the other image thresholding techniques. After the compressed histogram of an image is obtained using the
basic JND histogram algorithm, the agglomeration technique can further be used to reduce the number of colors
by combining the smaller segments (less than .1% of the image size) with similar colored larger segments. This
stimulates the process of merging of small left over segments with larger similar segments.
VI. RESULTS
Table of parameters
Result computed is analysed in terms of three parameters namely-
DSC (Dice Similarity Coefficient) is a measurement of spatial overlap used widely for comparing segmentation
results, which can have a value ranging from zero to one. Precision and recall is propotional to the total number
of unmatched pixels between two segmentations S1 and S2. Precision (P), defined as the fraction of detections
that are true boundaries and Recall (R), given by the fraction of true boundaries that are detected. Thus, Precision
quantifies the amount of noise in the output of a detector, while Recall quantifies the amount of ground-truth
detected.

6
Parameter Watershed
Algorithm
Color
Histogram
Proposed
Method
DSC 0.701595 0.413567 0.964047
Precision 0.82432 0.748208 0.897597
Recall 0.556199 0.494198 0.663132
Table 1. Comparative Result
Fig 5(a)Original image (b)Segmented image
VII. CONCLUSION
Image segmentation is a vital step in image processing. The purpose of image segmentation is to divide the
original image into homogeneous regions. There are several approaches to perform this task such as Edge-based,
Clustering-based, Region-based, etc. In this work the watersheds transformation was selected as a particular
region-based approach method to do the segmentation. The immersion process of the watersheds transformation
is a fast and powerful algorithm to produce the segmented image. The transformation needs pre-processing and
post-processing for embedded problems.
The evaluation of the segmentation is done by comparing the each object in true segmentation with the object in
marker-controlled watershed segmentation or proposed method. Conclusion is that the proposed method
enhances the result of marker-controlled watershed for degraded images. Here the enhancement is in terms of
robustness i.e. the outcome and accuracy. For example if watershed algorithm identifies three objects from an
image containing six objects. Then proposed system will try to identify four or five object from the same image
containing the six objects. The restoration and segmentation quality of the proposed method was visually and
quantitatively better than those of the other algorithms. The segmented output is more pleasing without
over-segmentation and the algorithm is also less time complex than other traditional algorithms.
VIII. FUTURE SCOPE
The domain of image segmentation is still immature. Many unsettled problems need to be defined and solved in
this area. It is firmly believed that this domain will greatly advance in the future
Following are the suggestions for future scope.
1. Improve the speed by using parallel processing either on clusters system or multiprocessor
system.
2. Merge this technique with some other technique to get the better results.
3. Various soft computing techniques can be used to have better result.
REFERENCES
[1] T. Malisiewicz and A. A. Efros, “Improving spatial support for objects via multiple segmentations,” in Proc.
BMVC, 2007
[2] Beucher, S.The Watershed Transform Applied to Image Segmentation, Proceedings of the P.Pfefferkon
Conference on Signal and Image Processing in Microscopy and MicroAnalysis,pp.299-314,September 1991.
[3] H. Digabel and C. Lantuéjoul, “Iterative algorithms,” Proc. 2nd
European Symp. Quantitive Analysis of
Microstructures in Material Science, Biology and Medicine, October 1977.
[4] F. Meyer, S. Beucher. Morphological Segmen- tation. Journal of Visual Communication and Image
Representation, 1(1):21-45, Sept. 1990.
[5] Luc Vincent and Pierre Soille, “Watersheds in digital space: An efficient algorithm based on immersion
simulations,” IEEE Tran. on Pattern Recognition and Machine Intelligence, Vol. 13, No. 6, June 1991.

7
[6] F. Meyer, S. Beucher. Morphological Segmentation. Journal of Visual Communication and Image
Representation, 1(1):21-45, Sept. 1990.
[7] R. Lotufo,W. Silva, Minimal set of markers for the watershed transform, in: Proc. ISMM 2002.
[8] F. Meyer, P. Maragos. Multiscale Morphological Segmentations Based onWatershed, Flooding, and Eikonal
PDE. Proc. Int’l Conf. on Scale-Space Theories in Computer Vision (SCALE-SPACE'99), Corfu,Greece, Sept.
1999.
[9] Serge Beucher, “The watershed transformation page: Image segmentation and mathematical
morphology,” cmm.ensmp.fr/~beucher/wtshed.html.
[10] Hao Wei, Zheng Sheng, Ye Shu-zhi, “One improved watershed transform for medical image segmentation”,
Computer Application and System Modeling (ICCASM), International Conference, 2010.
[11] Zhang Gui-Mei , Zhou Ming-Ming, Chu Jun, Miao Jun , “Labeling watershed algorithm based on
morphological reconstruction in color space”, Haptic Audio Visual Environments and Games (HAVE), IEEE
International Workshop, 2011.
[12] Zhonglin Xia ,Danfeng Hu, Xueyan Hu,Wei Xie, Qianqing Qin “Application of an improved watershed
algorithm in the insulator contamination Monitoring”, IEEE, 2011.
[13] Jie Chen, Meng Lei, Yao Fan, Yi gao, “Research on an improved Watershed algorithm to Image
segmentation”, The 7th International Conference on Computer Science & Education (ICCSE 2012), 2012.
[14] Shu-Yuan Shang, Ya-Xia Liu, Fei Meng, “An Improved Method Of Watershed Transform on Image of
Cashmere and Wool Fibre”, Proceedings of the International Conference on Machine Learning and Cybernetics,
Xian, 2012

This academic article was published by The International Institute for Science,
Technology and Education (IISTE). The IISTE is a pioneer in the Open Access
Publishing service based in the U.S. and Europe. The aim of the institute is
Accelerating Global Knowledge Sharing.
More information about the publisher can be found in the IISTE’s homepage:
http://www.iiste.org
CALL FOR PAPERS
The IISTE is currently hosting more than 30 peer-reviewed academic journals and
collaborating with academic institutions around the world. There’s no deadline for
submission. Prospective authors of IISTE journals can find the submission
instruction on the following page: http://www.iiste.org/Journals/
The IISTE editorial team promises to the review and publish all the qualified
submissions in a fast manner. All the journals articles are available online to the
readers all over the world without financial, legal, or technical barriers other than
those inseparable from gaining access to the internet itself. Printed version of the
journals is also available upon request of readers and authors.
IISTE Knowledge Sharing Partners
EBSCO, Index Copernicus, Ulrich's Periodicals Directory, JournalTOCS, PKP Open
Archives Harvester, Bielefeld Academic Search Engine, Elektronische
Zeitschriftenbibliothek EZB, Open J-Gate, OCLC WorldCat, Universe Digtial
Library , NewJour, Google Scholar

An efficient image segmentation approach through enhanced watershed algorithm

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to An efficient image segmentation approach through enhanced watershed algorithm (20)

More from Alexander Decker (20)

Recently uploaded (20)

An efficient image segmentation approach through enhanced watershed algorithm