EDGE DETECTION OF MICROSCOPIC IMAGE

http://www.iaeme.com/IJECET/index.asp 1 editor@iaeme.com
International Journal of Electronics and Communication Engineering & Technology
(IJECET)
Volume 7, Issue 3, May–June 2016, pp. 01–10 Article ID: IJECET_07_03_001
Available online at
http://www.iaeme.com/IJECET/issues.asp?JType=IJECET&VType=7&IType=3
Journal Impact Factor (2016): 8.2691 (Calculated by GISI) www.jifactor.com
ISSN Print: 0976-6464 and ISSN Online: 0976-6472
© IAEME Publication
EDGE DETECTION OF MICROSCOPIC
IMAGE
Bhupendra Fataniya, Mekhala Kar, Grishma Joshi, Dr. Tanish Zaveri, Dr.
Sanjeev Acharya
Institute of Technology, Nirma University, Ahmedabad, India
ABSTRACT
Edge detection of herbal plants is a set of mathematical methods which
aim at identifying points in a digital image at which the image brightness
changes sharply and has discontinuities. They are defined as the set of curved
line segments termed edges. Effective edge detection for microscopic image of
herbal plant is proposed through this paper which compares the edge detected
images and then performs further segmentation. Comparison between Sobel
operator, Prewitt, Canny and Robert cross operators is performed. Our
method after efficient edge detection performs Gabor filter and K-means
clustering to procure a better image. It is then subjected to further
segmentation. Experimental methods in our proposed algorithm show that our
method achieves a better edge detection as compared to other edge detector
operators. Our proposed algorithm provides the maximum PSNR value of
43.684 amongst the other commercial edge detection operators.
Key words: Microscopic Images, Gaussian Blur, Edge Detection, Image
Fusion, Clustering.
Cite this Article: Bhupendra Fataniya, Mekhala Kar, Grishma Joshi, Dr.
Tanish Zaveri and Dr. Sanjeev Acharya. Edge Detection of Microscopic
Image, International Journal of Electronics and Communication Engineering
& Technology, 7(3), 2016, pp. 01–10.
http://www.iaeme.com/IJECET/issues.asp?JType=IJECET&VType=7&IType=3
1. INTRODUCTION
Microscopic image processing is a broad term that covers the use of digital image
processing techniques to process, analyse and present images obtained from a
microscope. Such processing is now commonplace in a number of diverse fields such
as medicine, biological research, cancer research, drug testing, metallurgy, etc.
Microscopic image analysis is used in many fields of technology and physics. Typical
resolution may be of 1024x1024 pixels while others have a resolution of just 50x50
pixels. An image may be defined as a two-dimensional function, f(x, y), where x and
y are spatial (plane) coordinates, and the amplitude off at any pair of coordinates (x,

Bhupendra Fataniya, Mekhala Kar, Grishma Joshi, Dr. Tanish Zaveri and Dr. Sanjeev
Acharya
y) is called the intensity or grey level of the image at that point. When x, y, and the
amplitude values of f are all finite, discrete quantities, we call the image a digital
image. The field of digital image processing refers to processing digital images by
means of a digital computer. Note that a digital image is composed of a finite number
of elements, each of which has a particular location and value [1]
. These elements are
referred to as picture elements, image elements and pixels.
Our paper focuses on the implementation of the paper by authors Jun Li, Yuxi
Song, Yaoli Li, Shaoqin Cai and Zehong Yang with certain. The authors of this paper
mainly focused on the necessity of the identification of the herbal plants automatically
and to determine the region of interest. [2]
 They have applied an automatic segmentation technique based on the image
proposed.
 Firstly the image gradient is obtained and image fusion is performed to smoothen it.
 Further the Gabor feature is extracted and a k-means clustering is performed keeping
the value of k=2 for classifying it to background and foreground.
 Since the image featured extracts were not coming that accurately, so we had to make
modification in the procedure.
 We further performed segmentation involving the morphological procedure
consisting of erosion and dilation.[2]
The paper here focuses to develop an efficient automatic system to determine the
region of interest out of the given image. The best proposed methodology is the
automatic segmentation technique. In this paper we have mainly focused to achieve a
better output from the edge detection technique by comparing the various edge
detection operators. Basically we are performing this experiment to achieve the
automatic segmentation of the digital image through the edge detection technique. In
the section 2 we have explored all the concepts of the edge detection techniques.
Followed by that, we have discussed the technique of image fusion in the section 3.
Then in the section 4 and 5 we have performed the Gabor filter technique and image
feature extraction. In the section 6 we have detailed our experiment with the K-means
clustering procedure in order to obtain a better edge detected image amongst the other
commercial edge detected operators. Based on this methodology we have performed
our experiments on the herbal medicinal plants of Liquorice and Rhubarb.
(a) (b)
Figure 1 Original image of (a) Liquorice and (b) Rhubarb
The Liquorice plant is basically a fiber cell structure and the rhubarb is a rosette
crystal structure shown in Figure 1. From the original images we can clearly see the
non-uniformities present in the images. The image obtained is not that clear and has
many noises present.

Edge Detection of Microscopic Image
2. EDGE DETECTION
The edge is the set of those pixels whose grey have a step and rooftop change and it
exists between the object and the background, object and object, region and region
and between element and element. Edge always dwells in two neighboring areas
having different grey level. It is the result of grey level being discontinuous. Edge
detection is a kind of method of image segmentation based on range non-continuity.
When image is acquired, the factors such as the projection, mix, aberrance and noise
are produced. These factors bring on image feature’s blur and distortion, consequently
it is very difficult to extract image feature[4]
. Moreover, due to such factors it is also
difficult to detect edge. In order to gain more legible image outline, firstly the
acquired image is filtered and de-noised. In the process of de-noising, Gaussian blur is
used. There are three criterions relevant to edge detector performance:
 Low error rate
 Edge points to be well localized
 Circumvent the possibility of multiple responses to a single edge
Edge detection methods are based on difference operation and used widely in
image processing. It could detect variation of grey levels but sensitive to noise. The
edges of the images are the places where grey levels vary rapidly. The derivatives of
grey levels are used to detect the image edges in traditional methods [5]
. The
differences of grey levels can be used to obtain in the wider range of image so that
they are not suitable for detecting edge accurately.
The image detection on the acquired digital image of the herbal plants is done through the
edge detection technique. In this section we have experimented with the common edge
detection operators mainly Sobel, Canny and Prewitt. Through certain edge detection
parameters we have compared these three and realized that sobel operator is not alone going
to give an efficient output so we have designed the procedure in such a way that all the
modifications have been incorporated after the sobel detected output and we have further
developed a better technique to find out an efficient image. Our methodology is briefly
described under the following sections and it’s diagrammatically described through the block
diagram in Figure 2.
Figure 2 Block Diagram of our Approach
Image
Acquisition
Gray Scaling
Gaussian Filter Edge Detection
Image Fusion Gabor Filter
Feature
Extraction
K-Means
Clustering
New edge
detected image

Acharya
2.1. Sobel Operator
The Sobel operator applied on a digital image in gray-scale, calculates the gradient of
the intensity of brightness of each pixel, giving the direction of the possible increase
of black to white, in addition calculates the amount of change in that direction.
Through the image shown in Figure 3(a), the similar analogy is predicted which gives
us the appropriate amount of edge detection through this operator. The Sobel operator
performs the 2D spatial gradient measurement on an image. Typically it is used to
find the approximate absolute gradient measurement at each point in an input gray-
scale image. The Sobel edges detector uses a pair of 3x3 convolution masks, one
estimating the gradient in the x-direction (columns) and the other estimating the
gradient in the y-direction (rows). Sobel operator evaluates on the basis of the
parameters low threshold (30%, 50%) and high threshold (70%, 80%, and 90%) for
hysteresis [6].
2.2. Prewitt Operator
Prewitt operator reaches extreme detection edge in the edge and makes noise smooth
by gray scale difference of adjacent pixels from top to bottom, left and right. Prewitt
operator has eight templates corresponded with boundary. Every template makes a
special edge orientation. The maximum of all the eight direction are used to export of
edge amplitude image. Through the image shown in Figure 3(b), the similar analogy
is predicted which gives us the appropriate amount of edge detection through this
operator. Prewitt operator carries out edge detection to an image in 8 directions, and
let the maximum direction response be the edge of the edge magnitude image, it also
has the smoothing effect on the noises. The Prewitt operator is a method which uses
airspace differential convolution or an operation similar to convolution. The substance
is using a partial differential operator to do convolution to the image, and then using
the amplitude extremist of the first derivative or the amplitude zero-crossing point of
the second derivative to detect the edge. Prewitt operator can easily detect the noise
signals and have a certain degree of smoothing effect on the image [7]
. The detected
result will amplify the noise signals and blur the useful signals of the edge. This will
affect the image detection results significantly.
2.3. Canny Operator
Canny operator algorithm just uses the fixed high and low thresholds to extract edge,
without adaptively for different images and cannot eliminate noise obstruction on the
same time; there will be unreal edge detected and some weak edge lost for slow
change of gray scale. As shown in Figure 3(c), the similar analogy is predicted which
gives us the appropriate amount of edge detection through this operator. The canny
algorithm consists of three criterion of the edge detection algorithm:
 Criterion of high SNR for better edge detection
 Criterion of high positioning accuracy
 The criterion of singleness edge response
The error is the presence of the noise in the image hence using the Gaussian filter
to smooth the noise so that the edges are not weakened and lose their edge pixel value
[8]
. Sometimes even the single pixel value grade is not obtained. To improve the canny
detector instead of using the Gaussian filter, using the adaptive filter can be used. But
this too involves heavy calculation hence it is usually avoided.

(a) (b) (c)
Figure 3 Comparison of images (a) Sobel Operator (b) Prewitt Operator (c) Canny Operator
3. IMAGE FUSION
Image fusion is the process of combining of multiple images of the same scene
obtained from different sensors into a single image. The data from the images may
have different spatial or spectral resolutions, may come from different imaging
modalities, or may come from sensors which measure fundamentally different
attributes of the scene. The goal of image fusion is to effectively integrate similar and
dissimilar information obtained from the multiple images to form a new image which
provides a more informative description of the scene. Image fusion has one goal
which is to combine different images in order to improve the segmentation process
[10]
. Image fusion at the pixel level means fusion at the lowest processing level
referring to the merging of measured physical parameters as shown in Figure 4.
Figure 4 Fused image of the edge detected and gray scaled image
4. GABOR FILTER
Performances of different filters are compared for the intensity profile. The energy
profile is obtained using the odd/even Gabor and log Gabor filters. The edges
delimitating the bar is not well detected and hence a quadrature pair of Gaussian and
Gabor filters are used. Gabor transform is a popular procedure in frequency domain
and is used for solving image segmentation [3]
. With Gabor filters of multi-scale we
can measure the distance of two repeated elements. And with Gabor filters of multi-
angles we can measure the direction of each image obtained. A Gabor ﬁlter is
essentially a sinusoid modulated by a Gaussian function.

Acharya
(1)
where
λ
5. FEATURE EXTRACTION
Feature extraction refers to extract the information of image using computers to
determine whether each image pixel is the same type of image features. It is a key part
in computer vision and image processing field. Color features, text features, shape
features, and characteristics of spatial relations are commonly used image features.
We get M*N of images with each images width W and height H. We get the first
pixel value of each image to form a vector:
F(1,1) = (GaborImage1(1,1), GaborImage2(1,1), aborImageMN(1,1)) (2)
In the same way we get vector F(1,2), F(W,H). Now we can represent the value of
each pixel by a vector:
F(i,j) = (GaborImage1(i,j), GaborImage2(i,j), GaborImageMN(i,j)) (1≤ i ≤ W, 1 ≤
j ≤ H) (3)
6. K–MEANS CLUSTERING
Clustering is an important area of data mining which are placed between statistics and
informatics. Clustering is the process of grouping similar objects in such a way that
two objects from the same cluster are more similar than two objects from different
clusters. The clustering process may result in different grouping of a data set,
depending on the specific criterion used for clustering. According to the method
adopted to define clusters, the algorithms can be broadly classified into Partition
Clustering, Hierarchical Clustering, Density based Clustering, and Grid based
Clustering, Statistical Clustering, Fuzzy Clustering, Rough, Ant Colony Optimization
and Neural Network Clustering. There are different ways of classifying the cluster
analysis methods. Partitioning methods optimizes the assignment of the objects into a
certain number of clusters, and methods of hierarchical cluster analysis with graphical
outputs which make assignment of objects into different numbers of clusters possible.
In the partitioning methods, k-centroids and k-medoids methods are used for
disjunctive clustering. It is based on initial assignment of the objects into k clusters.
For this purpose, k initial centroids are selected for the clusters. Different approaches
are applied for selection of the initial centroids; for example, the first k objects can be
used. After that, the distances of each object from all centers are calculated. The
object is assigned to the closest centroid. The elements of the new centroids are
computed using the average values of individual variables. Then the distances of each
object from all centroids are calculated again. If an object is closer to the centroid of
any other cluster, it is moved to that cluster. This process is repeated until all objects
will be moved. In the k-medoids method, a certain vector of observations is taken for

the center of the cluster [11]
. The K-means algorithm is a widely used centroid based
partitioned clustering algorithm. K–means is a popular partitioned clustering method.
It is also known as generalized Lloyd algorithms (GLA) in which Euclidean distances
are used to measure the dissimilarity between data point and the cluster
representatives. The well-known k-means algorithm suffers the demerit that its
performance is dependent on the choice of the initial clusters and the instance order
[12]
. Using this method, the output of Liquorice plant is shown in Figure 5.
Figure 5 Clustered image of Liquorice
7. EXPERIMENTAL RESULTS
This paper focuses to achieve the most efficient edge detection of the image. We have
successfully performed the edge detection through the three most efficient operators
being Sobel, Prewitt and Canny.
These operators are compared and distinguished on the basis of certain parameters
being the PSNR (peak signal to noise ratio), MAE (mean absolute error), and IMMSE
(mean square error).
Table 1 Comparison of parameters of edge detection operators for Liquorice
Edge Operator PSNR MAE IMMSE
Sobel 43.2093 0.0121 2.0938e+04
Prewitt 43.2093 0.0123 2.0938e+04
Canny 43.2069 0.0566 2.0926e+04
Proposed Methodology 43.6842 27.8552 2.3357e+04
The mean square error and the peak SNR are the two error metrics used to
compare image compression quality [9]
. The MSE represents the cumulative squared
error between the compressed and the original image, whereas the PSNR represents
the peak error.
These operators are compared and distinguished on the basis of certain parameters
being the PSNR (peak signal to noise ratio), MAE (mean absolute error), and IMMSE
(mean square error). The mean square error and the peak SNR are the two error
metrics used to compare image compression quality [9]
. The MSE represents the
cumulative squared error between the compressed and the original image, whereas the
PSNR represents the peak error.
From Table 1, we can clearly detect that our proposed algorithm gives us the
desired output values and it’s the most efficient out of all the other commercial edge
detection operators. The desired range of the PSNR value of any processed image is
in the range of 30dB to 50dB.

Acharya
The entire methodology has been experimented for both the Liquorice and
Rhubarb plant and the results have been shown and compared in Figure 6.
(a)
(b)
(c)
(d)
(e)
(f)
(g) (h)
(i) (j)
Figure 6 (a) and (b) are original images of Liquorice and Rhubarb; (c) and (d) are edge
detected image using Sobel operator; (e) and (f) using Prewitt operator; (g) and (h) using
Canny operator; (i) and (j) are new edge detected image using proposed algorithm.

8. CONCLUSION
This paper proposes an effective automatic segmentation. To pursue an image of
interest firstly the image is acquired and it is gray-scaled. Further we have applied
Gaussian filter to remove the non-uniformities and noise from the image. Then the
image feature is extracted and clustered. Then the image is set up to gather the energy
equations. We have successfully compared the edge detection techniques based on
various parameters and have realized that our approach significantly helps to achieve
a better detected image. Our proposed algorithm provides the maximum value of
PSNR of 43.6842 out of the other edge detected operators. Further we have applied
morphological segmentation using two ways of erosion and dilation to classify the
image better and the same is shown in Figure 7. We are still facing the issues
regarding the resizing of the original image, and our segmented outputs are not
accurate enough.
(a) (b)
Figure 7 (a) Dilated image (b) Eroded image
So we are still in the working phase of developing a better approach of
segmentation and further will work on the advanced segmented methods like
watershed algorithm and Gabor-cut algorithm. In this paper as our morphological
segmentation outputs are not that clear so in our future work we have planned to
develop a better modified code to get an efficient output.
ACKNOWLEDGMENTS
We would like to extend our gratitude to the Pharmacy Department, NIRMA
University, for allowing us to perform all the required experiments. We would also
like to extend our appreciation to the GUJCOST Funding Agency for their support
and encouragement in the completion of our research.
REFERENCES
[1] Rafael c. Gonzalez, Richard E Woods, Digital Image Processing, Prientice Hall
publications, Upper Saddle River, New Jersey 07458
[2] Jun Li, Yixu Song, Yaoli Li, Shaoquin Cai, Zehong Yang, Automatic Target
Segmentation Based On Texture For Microscopic Images of Chinese Herbal
Plants, 978-1-4673-5534-6/13, Pek University,Beijing, China.
[3] Anil K.Jain, Nalini K Ratha, and Sridhar Lakshmanana, Object Detection Using
Gabor Fiters, 30(2), pp.295-309,1997,Michigan University, USA
[4] Renyan Zhang, Guoliang Zhao and Li Su, A new edge detection method in image
processing, 0-7803-9538-7/05/2005, Automation college, Harbin, China.

Acharya
[5] Jinyong Cheng, Ruojuan Xue, Wenpeng Lu, Ruixian Jia, Segmentation of
Medical Images with Canny Operator and GVF snake model, Vol 7th
, 978-1-424-
2114-5/08/China.
[6] Olivia Mendoza, Patricia Melin, Guillerno Licea, A new method for Edge
Detection in Image Processing Using Interval Type–2 Fuzzy logic, 0-7695-3032-
x/07 UABC university, Tijuana Institute of Technology.
[7] Wensho Gao, Lei Yang, Xiaoguang Zhang, Bin Zhou, Chunxi Ma, Based on
Soft-thrshold wavelet de-noising combining with Prewitt operator Edge detection
alogrithm, 978-1-4244-6370-1/2010, IEEE, Information engineering school of
communication university of China.
[8] Bing Wang, Shao Sheng Fan, An improved Canny Edge Detection Alogrithm,
978-0-7695-3881-5/2009, IEEE, Changsha University of Science and
Technology, China.
[9] Mike Heath, Sudeep Sarkar, Thomas Sanochi, Kevin Bowyer, Comparison of
Edge Detectors: a methodology and initial study, 1063-6919/96/University of
South Florida.
[10] Shahan Nercessian, Karen Panetta, Sos Agaian, An Edge-Based Approach To
Image Fusion of Images via Reconstruction Estimation, 978-1-4244-4179-
2/2009.IEEE, University of Texas.
[11] Abhay Kumar, Ramnish Sinha, Daya Shankar Verma, Satendra Singh, Modelling
Using K–Means Clustering Alogrithm, 978-1-4577-4/2012,IEEE,BITS,Ranchi.
[12] Ashok Kumar, Mc Locarine Charlet, Ummal Sariba Begum, Computational Time
Factor Analysis of K-Means Alogrithm on Actual and Transformed Data
Clustering, 978-1-4673-1039-0/2012 IEEE, Govt of Arts College, Trichy.
[13] A. Al-Marakeby. A Novel Method For Clumped Particles Separation In
Microscopic Images, International Journal of Computer Engineering &
Technology, 5(1), 2014, pp. 52–61.
[14] Benayad Nsiri, Salma Nagid and Najlae Idrissi. New Approach To Multispectral
Image Fusion Based on A Weighted Merge, International Journal of Electronics
and Communication Engineering & Technology, 4(1), 2013, pp. 25–34.
[15] Jaspreet Kaur and Chirag Sharma. Multimodality Medical Image Fusion Using
Improved Contourlet Transformation with Log Gabor Filters, International
Journal of Electronics and Communication Engineering & Technology, 4(2),
2013, pp. 383–389.
[16] Mohammad Awad, Kacem Chehdi, Ahmad Nasim, Enhancement of the
Segmentation Process of Muti componenet images using Fusion with Genetic
Alogrithm, 978-1-4244-2206-7/08, IEEE, Beirut American University, Lebanon.

EDGE DETECTION OF MICROSCOPIC IMAGE

Recommended

More Related Content

What's hot (19)

Similar to EDGE DETECTION OF MICROSCOPIC IMAGE (20)

More from IAEME Publication (20)

Recently uploaded (20)

EDGE DETECTION OF MICROSCOPIC IMAGE