IJETR031551
IJETR031551
IJETR031551
73 www.erpublication.org
Image Segmentation Watershed Process based on Progressive Median Filtering & Gradient Map
segmentation is in general successful, however, in some cases, number of initial classes in advance. Ming-Xin Zhang et.al [9]
the color and spatial information are not sufcient for superior proposed conventional image segmentation based on
segmentation. They used wavelet frames that provide rough-set theory. The new distance matrix is defined by using
translation invariant texture analysis. Their proposed method the vector angle and Euclidean distance. And then according
integrates additional texture feature to the color and spatial to this new distance matrix, the space binary matrixes of each
space of standard mean-shift segmentation algorithm. color component are calculated. Further, the color image
The seeded region growing (SRG) algorithm is one of the segmentation is implemented by selection of threshold values
simplest region-based segmentation methods. It performs a and region merging through introducing a histogram based on
segmentation of an image with examine the neighboring roughness. This proposed approach yields better
pixels of a set of points, known as seed points, and determine segmentation which is more intuitive to human vision
whether the pixels could be classified to the cluster of seed compare with the existing conventional image segmentation.
point or not [4]. Unfortunately, this method is a time Chunming Li et.al [10] introduced a new variation level set
consuming and initial seed point needs to be specified. formulation in which the regularity of the level set function is
Unlike most of supervised and region growing algorithms, intrinsically maintained during the level set evolution. Level
fast scanning algorithm do not need seed point. In brief, the set is derived as the gradient flow that minimizes energy
concept of fast scanning algorithm [5] is to scan from functional with a distance regularization term and an external
theupper-left corner to lower-right corner of the whole image energy that drives the motion at desired locations. The
and determine if we can merge the pixel into an existed distance regularization obtained by a potential function,
clustering. The merged criterion is based on our assigned which is able to maintain a desired shape of the level set
threshold. If the difference between the pixel value and the function, particularly a signed distance profile near the zero
average pixel value of the adjacent cluster is smaller than the level set. The distance regularization effect eliminates the
threshold, then this pixel can be merged into the cluster. need for re-initialization and thereby avoids its induced
Tang. J in 2010 proposed that proposes a color image numerical errors. Further, this framework simpler and more
segmentation method of automatic seed region growing on efficient finite difference scheme can be used to implement
basis of the region with the combination of the watershed the DRLSE formulation. Thus to address the limitation of the
algorithm with seed region growing algorithm which based on commonly used existing segmentation algorithms, we design
the traditional seed region growing algorithm [15]. an effective spatial domain segmentation framework based on
Lei Li et.al [6] proposed an improved method for color the watershed algorithm. In addition, the proposed system
segmentation which uses the FSVM (fuzzy support vector uses median filter in pre-processing stage to limit the over
machines) algorithm in the HSI (hue-saturation-intensity) segmentation of the algorithm. Then the marker contours are
color space. Their system offers better effect and lower generated as a map based on the sobel defined operator.
computational complexity on color image segmentation than
most of existing methods. It was noticed that HSI color III. WATERSHED ALGORITHM
system has a good capability of representing the colors of The watershed algorithm is more representative in the
human perception. Unfortunately, the conversions of color application of mathematical morphology theory for image
space causes loss of valuable and time consuming process on segmentation. Watershed algorithm is a region-based
the whole. Wen bing Tao et.al [7] introduced a novel segmentation techniques image that uses image morphology.
approach that provides effective and robust segmentation of The prime initiative of watershed segmentation algorithm is
color images. By incorporating the advantages of the mean based on the concept of topographic representation of image
shift (MS) segmentation and the normalized cut (Ncut) intensity. Marker is knowledge about the object based on
partitioning methods. In addition, they illustrate that the low application-oriented; it is selected by the operator manually
computational complexity makes their method feasible for or by automatic process. The objects can use watershed
real-time image segmentation processing. It also preprocesses algorithm to transform and develop regional growth after
an image by using the MS algorithm to form segmented maker. Meanwhile, Watershed segmentation also embodies
regions that preserve the desirable discontinuity other principal image segmentation methods including
characteristics of the image. Further, this proposed method discontinuity detection, thresholding and region processing.
allows a low-dimensional image clustering with significant Because of these factors, watershed segmentation displays
reduction of the complexity compared to conventional graph more effectiveness and stableness than other segmentation
partitioning methods that are directly applied to the image algorithms.
pixels. In addition, the image clustering using the segmented
regions reduces the sensitivity to noise and results in Geethalakshmi. S. N, and Jothi.T in 2010 works on
enhanced image segmentation performance. Medical imaging and analysis. Image segmentation and
Cheng-Wan An et.al [8] proposed an adaptive enhancement is one of the central research themes in medical
segmentation approach for color image in RGB space based image analysis and with the increasing number of imaging
on Rival Penalized Competitive Learning (RPCL). Initially, analysis and imaging studies, the necessity for automated
histograms of Red, Green, and Blue component colors are medical image segmentation methods are expanding. Their
segmented by adaptive threshold segmentation algorithm for research work focus on segmentation that combines neural
gray image. The some possible colors of original image are network with Watershed algorithm modified to use
specified through combining those components' segmented automatic optimal threshold selection algorithm. The
areas and false colors not appearing in image are removed. problem of over-segmentation is solved by using
Finally those possible colors are converged to the main actual preprocessing technique. Experimental results prove that the
colors of original image by RPCL. Then original image is
segmented by those learned centers without specifying the
74 www.erpublication.org
International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-3, Issue-3, March 2015
proposed method is efficient in segmenting medical images The prime objective of watershed algorithm is to determine
[16]. all of the watershed lines (the discontinuity in the image). The
Navon. E in 2004 proposed a method of image most intuitive way to explain watershed segmentation is the
segmentation which was based on adaptive local thresholds Immersion Approach [11]: imagine that a hole is drilled in
[17]. Their proposed algorithm divides the image into each minimum of the surface, and we flood water into
homogeneous regions by local thresholds. The number of different catchment basins from the holes. If the water of
thresholds and their values are adaptively derived by an different catchment basins is likely to merge due to further
automatic process, where local information is taken into immersion, a dam is built to prevent the merging. This
consideration. First, the watershed algorithm is applied. Its flooding process will eventually reach a stage when only the
results are used as an initialization for the next step, top of dam (the watershed lines) is visible above the water line.
which is iterative merging process. During the iterative Further, the much discussed issues surrounding the watershed
process, regions are merged and local thresholds are derived. algorithm are well explained using the figure 1 & 2. Figure 1,
The thresholds are determined one-by-one at different times shows the original image that would be processed using
during the merging process. Every threshold is calculated by watershed algorithm to obtain the segmented regions. The
local information on any region and its surroundings. figure 2 presents the over segmentation problem associated
Kishor Bhoyar & Omprakash Kakde in 2010 proposed a with the watershed algorithm. In order to limit this drawback,
new algorithm for Image segmentation which was based on we propose a combination of Median-filtering, soft
JND colour histogram. In their proposed model Histogram of thresholding and maker controlled pre-process.
the given color image is computed using JND color model
[18]. This samples each of the three axes of color space so
that just enough number of visually different color bins
(each bin containing visually similar colors) are obtained
without compromising the visual image content. The
number of histogram bins are further reduced using
agglomeration successively. This merges similar histogram
bins together based on a specific threshold in terms of JND.
This agglomerated histogram yields the final segmentation
based on similar colors.
P. S. Hiremath and Jagadeesh Pujari in 2010 proposed a
method for Image Retrieval based on Color, Texture and
Shape features using Image and its complement presents a
novel framework for combining all the three i.e. color, texture
and shape information, and achieve higher retrieval FIG.1 ORIGINAL IMAGE FLOWER
efficiency using image and its complement [19]. The
image and its complement are partitioned into
non-overlapping tiles of equal size. The features drawn from
conditional co-occurrence histograms between the image
tiles and corresponding complement tiles, in RGB color
space, serve as local descriptors of color and texture. This
local information is captured for two resolutions and two grid
layouts that provide different details of the same image.
The combination of the color and texture features between
image and its complement in conjunction with the shape
features provide a robust feature set for image retrieval.
Jeong. S in 2001 present a report on image retrieval, and
proposed a method of image retrieval based on Histogram
Based [20]. In his project, six histogram-based search Fig.2 Segmented regions using watershed algorithm
methods are investigated in two different color spaces, RGB
and HSV. Histogram search characterizes an image by its
color distribution, or histogram but the drawback of a global
histogram representation is that information about object
location, shape, and texture is discarded. Thus this project
showed that images retrieved by using the global color
histogram may not be semantically related even though they
share similar color distribution in some results. An image
retrieval demo system was built to make it easy to test the
retrieval performance and to expedite further algorithm
investigation. And six histogram-based image retrieval
methods in two color spaces were exhaustively compared by
providing precision vs. recall graphs for each image class and
for all test images. In general, histogram-based retrievals in
HSV color space showed better performance than in RGB Fig.3 Segmented regions using watershed algorithm after
color space. median filtering
75 www.erpublication.org
Image Segmentation Watershed Process based on Progressive Median Filtering & Gradient Map
A. Median Filtering
B. Soft Thresholding
76 www.erpublication.org
International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-3, Issue-3, March 2015
algorithm may also be applied on 24-bit, each color layer V. COMPUTER SIMULATION
images treated as an individual 8-bit image. Computer simulations were simulated using MATLAB
Pre-Processing Analysis. The image is converted into software package. Analysis was done using 100 color images
gray-scale image by combining RGB components with fixed of varying sizes, texture and contour. These images were
weights. The median filtering analysis of an image is taken using 2 digital cameras Nikon D100 and Canon EOS
considered as a smoothing distribution over the pixels values. Digital Rebel and modified in Photoshop to attain a smooth
Further, it provides a means to determine the information histogram. In addition, for testing the effectiveness and
content of the image but it would not take spatial factors into efficiency of the proposed system varying images with
consideration. Further, the thresholds should be chosen based different image features are employed Figure 7 shows the
on the regional peak values among the image regions. segmentation results of a flower image with objects having
Gradient Matrix. The gradient matrix is the key feature of weak edges. Figure 7(a) shows the input image to our model.
the proposed segmentation process as the efficiency of the The figure 7 (b) shows the segmentation result that is obtained
segmentation process depends on the how accurate the by the proposed watershed framework.
gradient matrix is defined. The aim is to retain as much as
edge detail information as possible in the gradient map. This
method has the advantage of not having to estimate any
parameters, as they are evaluated based on the pre-defined
sobel operator. The gradient is high at the borders of the
objects and low (mostly) inside the objects.
Soft Threshold based System. Soft threshold based
technique is one of the commonly used image processing
operation for digital filtering. The proposed system uses the a) b)
soft thresholding to evaluate each binary label with reference
to the input image and calculate the combined homogeneity of Fig.7 (a) Input Image flower (b) Objects detected based on
the region. In addition, we also use a global threshold function proposed segmentation algorithm
based on Otsu's method, which chooses the threshold to
minimize the intra-class variance of the black and white pixels.
The topographical distance function to obtain a watershed
segmentation which is formulated within the level set theory
is also obtained in this phase.
Watershed Algorithm System. Watershed segmentation
also embodies other principal image segmentation methods
including discontinuity detection, thresholding and region
processing. Because of these factors, watershed segmentation
displays more effectiveness and stableness than other a) b)
segmentation algorithms. Fig.8 (a) Input Image pears (b) Objects detected based on
Outputs: - Segmented Image. proposed segmentation algorithm
The Segmentation process is a straightforward process Figure 8 shows the segmentation results of a pears image
where all necessary parameters are dictated based on the with objects having weak edges and smooth texture. Figure
combination of median filtering, sobel operation and soft 8(a) shows the input image to our model. The figure 8 (b)
thresholding. Input image is segmented into various uniform shows the segmentation result that is obtained by the proposed
and non-uniform regions without any loss or integrity with watershed framework
reference to the input with highest efficiency and low time The simulation analysis show that the proposed algorithm is
complexity. superior to the existing supervised segmentation algorithms
and in some cases on synthetic images it segments similar to
the existing algorithms.
VI. CONCLUSION
In this paper, we introduced a watershed based segmentation
framework based on the combination of median filtering,
sobel operation and soft thresholding process. The median
filtering analysis offer a better smoothing means and thus
reducing the impact of over segmentation on the images due
to watershed algorithm. Soft thresholding that is carried based
on the regions maximum value to obtain binary segments of
various classes that boasted significantly the watershed
algorithm performance. The segmentation task was tasted on
various images and the experimental results obtained by the
segmentation process are more efficient and reliable when
Fig. 6 The basic components of the proposed watershed based compare to the current techniques. Simulation results and
segmentation framework. analysis also proved that the proposed algorithm shows good
77 www.erpublication.org
Image Segmentation Watershed Process based on Progressive Median Filtering & Gradient Map
performance in image segmentation without choosing the Priyanka shivhare, she received her degree in Electronics &
region of interest. Communication from RGPV in the year 2011, Currently she is pursuing her
M.TECH in Digital Communication from RDIST Bhopal..
78 www.erpublication.org