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

net/publication/321776293

A Comprehensive Analysis of Image Edge Detection Techniques

Article  in  International Journal of Multimedia and Ubiquitous Engineering · November 2017

DOI: 10.14257/ijmue.2017.12.11.01

CITATIONS READS

20 4,569

3 authors:

Mohd. Aquib Ansari Diksha Kurchaniya

Motilal Nehru National Institute of Technology Motilal Nehru National Institute of Technology
19 PUBLICATIONS   78 CITATIONS    8 PUBLICATIONS   42 CITATIONS   

SEE PROFILE SEE PROFILE

Manish Dixit
Madhav Institute of Technology & Science Gwalior
90 PUBLICATIONS   977 CITATIONS   

SEE PROFILE

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

content based image retrieval View project

An Enhanced Approach of CBIR using Gabor Wavelet and Edge Histogram Descriptor View project

All content following this page was uploaded by Diksha Kurchaniya on 06 July 2018.

The user has requested enhancement of the downloaded file.

 International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017), pp.1-12
http://dx.doi.org/10.14257/ijmue.2017.12.11.01

A Comprehensive Analysis of Image Edge Detection Techniques

1

Mohd. Aquib Ansari1, Diksha Kurchaniya2 and Manish Dixit3

1, 2, 3
Department of CSE & IT
1, 2, 3
Madhav Institute Technology & Science, Gwalior, M.P., India
1
[email protected], [email protected], 3dixitmitsgmail.com

Abstract
One of the important objectives of image processing is to interpret the content of image
efficiently and finds the meaningful and significant information from it. The much
awareness has been received from various researchers in the field of image
interpretation. One of the most severe step in image interpretation is to mine the edges
information from the image appropriately. Edges are the fundamental features of the
image and can be formed from the outlines of the object. Edge detection is generally used
in image analysis and processing. There are several types of algorithm to detect the
edges. In this paper, the comprehensive analysis is done on the several edge detection
techniques such as Prewitt, Sobel, Canny, Roberts and Laplacian of Gaussian. It is
experimentally observed that Canny edge detector is working well than others. This work
is implemented on Matlab R2015a.

Keywords: Image segmentation, Edge detection, Prewitt, Sobel, Laplacian of

Gaussian, Canny, Roberts, PSNR, SNR

1. Introduction
Image processing is a method of analyzing and manipulating the digital images with
the computer using mathematical operators. In image processing, the input is an image
and outcome may be either set of characteristics or set of the parameter of image or an
image. An image comprises various information like contour of the object, its orientation,
size and color. So, as to find the shape information of the object, the edges involving in
that object must be identified. Edge detection is a method to detect the occurrence of
edges and its locality which is created by sharp and abrupt variations in intensity
(brightness or color) of an image. The discontinuities of an image can be variation in
scene illumination, discontinue in scene, surface orientation, its depth and variation in
material properties. The objectives of edge detection are to detect the shape information
of the object and the reflectance in the image. Edge detection is the important step in
image analyzing and processing, computer vision, human vision, object detection and
pattern recognition. There are various edge detection techniques for detecting the edges.
The different edge detectors work differently. Means some edge detectors take more time
and detects more edges with respect to others. The edge detection in an image is rest on
intensity, illumination, objects, noise, blur [1] [2] [3].
In this paper, various techniques of edge detection are studied for identifying the edges
in an image and the comparative analysis is also performed among these techniques.

2. Edge Detection Techniques

The edge demonstration of the image decreases the amount of information to be
processed, which contains vital information about the object’s shape in an image.

Received (June 25, 2017), Review Result (October 30, 2017), Accepted (November 20, 2017)

ISSN: 1975-0080 IJMUE

Copyright ⓒ 2017 SERSC
International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

Edges are local variations in image intensity. The edge forms between the
boundaries of two regions. The main information can be mined from the edge.
Edge detection [2] [4] is a process to locate the edges that having good
orientation and it is an essential tool of image segmentation. Edge detection method
transforms the original image into edge image with the help of operators. It is a
well-known process for identifying the dis-continuities in intensity values. In the
process of edge detection, the image is inputted first and converts that image into
gray scale image. And then apply the edge detector to detect and extract the edges
present within an image as output.
The various techniques are available for detecting the edges information such as
Roberts, Prewitt, Sobel, Laplacian of Gaussian and Canny. These techniques are
described as follows.

2.1. Roberts Edge Detection

Lawrence Roberts has proposed the Roberts edge detection technique [1] [11] for
detecting the edges within an image in 1965. It is a simple and computationally
efficient approach. It measures the spatial gradient of an image. The pixel value at
that point in the resultant image characterizes estimated absolute magnitude value of
the spatial gradient of the inputted image at that point. It takes input image as gray
scale image and produces edges involving in that image. The main disadvantages of
this technique are that it can’t detect that type of edges which are multiplie s of 45
degrees and it is not symmetric. The Robert operator contains the pair of 2x2
convolution masks which are illustrated in Figure 1. One mask is just to other
rotated by 90 degrees.

Figure 1. Masks used for Roberts Edge Detection Technique

The partial derivative of Robert operator is given as follows :

(1)

(2)

This operator produces the position of edges more accurately, but it has the short
support of filters which causes vulnerability to noise.

3.2. Prewitt Edge Detection

Prewitt has proposed the Prewitt edge detection technique [1] [6] [9] in 1970. It is
a right algorithm to measure the magnitude and orientation of the ed ges. This
technique evaluates the edge directions directly with the maximum response from
the mask. It is having 8 directions. But, sometimes most direct directions
approximation are not much perfect. This Prewitt operator is just like a Sobel

2 Copyright ⓒ 2017 SERSC

 International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

operator and easy to implement than Sobel operator but it produces some times
noisier results. The pair of 3x3 convolution masks for 8 directions are illustrated in
Figure 2. One mask is just to other rotated by 90 degrees .

Figure 2. Masks used for Prewitt Edge Detection Technique

The organization of pixels about central pixel is as follows:

The partial derivatives of Prewitt operator are measured as:

(3)

(4)

The Prewitt edge detector is less vulnerable to noise because it differentiates in one
direction and make average in another direction

2.3. Sobel Edge Detection

Irwin Sobel has proposed the Sobel edge detection technique [1] [11] [10] in 1970. The
Sobel kernel depends on the central difference, but while averaging it gives more weight
to central pixel. One of the advantages of Sobel kernel over Prewit kernel is that it has
better noise suppression characteristics. The Sobel edge detection method contains the
pair of 3x3 convolution masks illustrated in Figure 3. One mask is just to other rotated by
90 degrees. This mask can deal with the edges which are running 45 degrees to the pixel
grid. This mask can be put on distinctly to the input image to give gradient components in
every orientation.

Figure 3. Masks used for Sobel Edge Detection Technique

Copyright ⓒ 2017 SERSC 3

International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

The partial derivatives for Sobel operator are as follows:

(5)

(6)

The Gradient magnitude is as follows:

(7)

The Orientation angle is measured as follows:

(8)

2.4. Laplacian of Gaussian (LOG) Edge Detection

Marr has introduced the Laplacian of Gaussian (LOG) technique [11] [12] in (1982).
LOG is based on second order derivative. Which stated as:

(9)
LOG smoothes the image first then calculate Laplacian. This process produces the
double edge image. It locates edges then search the zero crossing between the double
edges. The LOG edge detection method contains the pair of 3x3 convolution mask which
is illustrated in Figure 4.

Figure 4. Masks used for LOG Edge Detection Technique

2.5. Canny Edge Detection

John Canny introduced the canny edge detection technique [1] [5] [11] at MIT in 1983.
It is the standard, powerful and usually used edge detection method. It separates the noise
from the image before extracting edges. Canny is a better method for extracting the edges
than other existing methods and produces the good result. The Canny operator can control
a number of details of edge image and can suppress the noise efficiently.
This method follows following steps:
1) For smoothing the image, the Gaussian filter is used with the identified value of
sigma which reduces noise.
2) At each point, the edge direction and local gradient are calculated. According to
edge point, it is the point with locally maximum strength in the gradient direction.
3) The edge point increases ridges in the gradient image magnitude. In this
algorithm, top of these ridges are considered and gives zero value to all pixels that are not

4 Copyright ⓒ 2017 SERSC

 International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

on the ridges top. Then, as output, a thin line is produced. This process is called non-
maximum suppression. Then, hysteresis thresholding is used here to threshold the ridges
pixel. It has two threshold values such as T1 and T2. Case1: if T1>T2, ridges pixel value
is higher than threshold T2, shows strong edge pixels. Case 2 if T2>T1, if T1<T2, ridges
pixel value is lesser than threshold T2, shows weak edge pixel.
4) At last, the edge linking process executes by integrating the weak pixels that
having 8-pixel connectivity to strong pixels.
The canny edge detection method contains the pair of 3x3 convolution mask shown in
Figure 5.

Figure 5. Masks used for Canny Edge Detection Technique

2.6. Parameter Used For Comparison

In this paper, PSNR and MSE are used to measure the performance of each edge
detection technique.

2.6.1. Peak Signal to Noise Ratio (PSNR)

PSNR [7] [11] [12] [13] [14] [15] is measured as the ratio of maximum possible power
and occurring noise that can disturb the representation of the image. PSNR is measured in
decimal scale. To calculate the quality reconstruction of an image, PSNR is used
commonly by the various researchers. It is a case where original data is treated as signal
and occurring error is treated as noise. The maximum value of PSNR shows high image
quality. The PSNR can be expressed as follows:

(10)

Where MAX = Maximum pixel value of image and image is represented by 8

bit/sample.

2.6.2. Mean Square Error (MSE)

MSE [8] [11] [12] [13] measures the true pixel value of usual information with the
degraded picture, it is used for the realistic purpose. Generally, MSE is calculated as the
average of square of the error between the genuine image and noisy image. The error can
me estimated as the difference between genuine image and degraded image. Here, lesser
value of MSE shows high and best quality image.

(11)

Copyright ⓒ 2017 SERSC 5

International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

3. Experimental Results
The experiment is done on Matlab R2015a and tested with the Sunflower and Face
image. To extract the clean edges map by using the principle edge feature of image is our
objective. Here various experiments, to detect and extract the edges, have been done with
the noise and without noise environment. In Figure 6 and Figure 10, the results by
applying various edge detection techniques on original image are displayed. In Figure 7
and Figure 11, the results by applying various edge detection techniques on Salt and
Pepper noise effected image are shown. In Figure 8 and Figure 12, the results by applying
various edge detection techniques on Gaussian noise effected image are shown. In Figure
9 and Figure 13, the results of Speckle noise effected image by applying various edge
detection techniques are shown.

Figure 6. Original Sunflower Image with the Outcome of Different Edge

Detection Techniques

Figure 7. Salt & Pepper Noise Effected Sunflower Image with the Outcome
of Different Edge Detection Techniques

6 Copyright ⓒ 2017 SERSC

 International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

Figure 8. Gaussian Noise Effected Sunflower Image with the Outcome of

Different Edge Detection Techniques

Figure 9. Speckle Noise Effected Sunflower Image with the Outcome of

Different Edge Detection Techniques

Figure 10. Original Face Image with the Outcome of Different Edge
Detection Techniques

Copyright ⓒ 2017 SERSC 7

International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

Figure 11. Salt & Pepper Noise Effected Face Image with the Outcome of
Different Edge Detection Techniques

Figure 12. Gaussian Noise Effected Sunflower Image with the Outcome of
Different Edge Detection Techniques

Figure 13. Speckle Noise Effected Sunflower Image with the Outcome of
Different Edge Detection Techniques

8 Copyright ⓒ 2017 SERSC

 International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

Here, Table 1 illustrated the PSNR and MSE value for the results of various edge
detection techniques on Sunflower Image in present of noise and no noise. Table 2
illustrated the PSNR and MSE value for the results of various edge detection techniques
on Face Image in present of noise and no noise. Table 3 describes the advantages and
disadvantages of the edge detection techniques.

Table 1. PSNR and MSE Value for Various Edge Detection Techniques on
Sunflower Image
Edge Original Salt & Pepper Gaussian Noise Speckle Noise
detection Sunflower Image Noise Effected Effected Effected
Techniques Sunflower Image Sunflower Image Sunflower Image
PSNR MSE PSNR MSE PSNR MSE PSNR MSE
Roberts 9.0044 8.177e+03 8.6962 8.779e+03 8.7929 8.586e+03 8.9127 8.352e+03
Sobel 9.0042 8.177e+03 8.6965 8.778e+03 8.7937 8.584e+03 8.9137 8.350e+03
Prewitt 9.0042 8.178e+03 8.6965 8.778e+03 8.7937 8.584e+03 8.9135 8.350e+03
LOG 9.0072 8.178e+03 8.7024 8.766e+03 8.8001 8.571e+03 8.9184 8.341e+03
Canny 9.0098 8.167e+03 8.7087 8.754e+03 8.8077 8.556e+03 8.9236 8.331e+03

Table 2. PSNR and MSE Value for Various Edge Detection Techniques on
Face Image
Edge Original Face Salt & Pepper Gaussian Noise Speckle Noise
detection Image Noise Effected Effected Face Effected Face
Techniques Face Image Image Image
PSNR MSE PSNR MSE PSNR MSE PSNR MSE
Roberts 5.6363 1.776e+04 5.6017 1.791e+04 5.5976 1.801e+04 5.5908 1.794e+04
Sobel 5.6364 1.776e+04 5.6024 1.791e+04 5.5982 1.801e+04 5.5918 1.794e+04
Prewitt 5.6365 1.776e+04 5.6023 1.791e+04 5.5982 1.801e+04 5.5918 1.794e+04
LOG 5.6383 1.775e+04 5.6057 1.770e+04 5.6015 1.800e+04 5.5958 1.792e+04
Canny 5.6395 1.774e+04 5.6080 1.779e+04 5.6038 1.799e+04 5.5992 1.791e+04

Table 3. Advantage and Disadvantage of Different Edge Detection

Techniques
Edge Detection Advantage Disadvantage
Techniques
Roberts (Based on First Produces more accurate Not reliable to extract the
Order Derivatives) position of edges edge in presence of noise
Sobel (Based on First Good noise suppression Produce moderate result
Order Derivatives) characteristics
Prewitt (Based on First Masks have longer support, Produce sometimes
Order Derivatives) Prewitt is less vulnerable to noisier result
noise
LOG (Based on Second Having fixed characteristics in Sensitive to noise,
Order Derivatives) all the directions, detects good generate closed and non-
edges and its orientations realistic contour
Canny (Based on Better detection specially in Complex, time
Second Order noise condition consuming, false zero
Derivatives) crossing

Copyright ⓒ 2017 SERSC 9

International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

5. Conclusion
In this paper, various edge detection techniques are studied and compared. After
the experimental analysis, it is found that the second order derivatives (Canny and
Log) is working well in comparison to first order derivatives (Sobel, Prewitt and
Roberts). The Log and Canny edge detection method producing good results for
image quality and visual perception. Since, Log edge detection technique is
vulnerable to noise. So, it is not providing the better results than canny edge
detection technique in presence of noise. Hence, it is experimentally proved that the
canny edge detector is a better edge detector technique of forming the edges for
inner as well as outer lines of the object. It has more good resistance to noise than
Roberts, Prewitt, Sobel and Log edge detection technique. Here, Sobel edge
detection technique proves better for discovering better outer lines (continuous
boundary) only of an object. As the future work, we can design the new filter over
the limitation to get better image quality so that the image can be enhanced by
reducing the noise.

References
[1] S. Das, “Comparison of various edge detection technique”, International Journal of Signal Processing,
Image Processing and Pattern Recognition, vol.9, no.2, (2016), pp.143-158.
[2] E. Nadernejad, S. Sharifzadeh and H. Hassanpour, “Edge Detection Techniques Evaluations and
Comparisons”, Applied Mathematical Sciences, vol. 2, no. 31, (2008), pp. 1507 – 1520.
[3] R. Maini and H. Agrawal, “Study and Comparison of Various Image Edge Detection Techniques”,
International Journal of Image Processing (IJIP), vol. 3, issue 1, pp. 1-12
[4] P. P. Acharjya, R. Das and D. Ghoshal, “Study and Comparison of Different Edge Detectors for Image
Segmentation”, Global Journal of Computer Science and Technology Graphics & Vision, (2012), vol.
12, issue 13, version 1.0.
[5] M. Juneja and P. Singh Sandhu, “Performance Evaluation of Edge Detection Techniques for Images in
Spatial Domain”, International Journal of Computer Theory and Engineering, vol. 1, no.5, (2009), pp.
614-621.
[6] R. Muthukrishnan and M. Radha, “Edge Detection Techniques for Image Segmentation”, International
Journal of Computer Science & Information Technology (IJCSIT), vol 3, no 6, (2011), pp.-259-267.
[7] L. Bin and M. S. Yeganeh, “Comparison for Image Edge Detection Algorithms”, IOSR Journal of
Computer Engineering (IOSRJCE), ISSN: 2278-0661, vol. 2, issue 6, (2012), pp. 01-04.
[8] H.S. Bhadauria, A. Singh and A. Kumar, “Comparison between Various Edge Detection Methods on
Satellite Image”, International Journal of Emerging Technology and Advanced Engineering, vol. 3, issue
6, (2013), pp.-324-328.
[9] S. Kaur and I. Singh, “Comparison between Edge Detection Techniques”, International Journal of
Computer Applications, vol. 145, no.15, (2016), pp. 15-18.
[10] V. Saini and R. Garg, “A Comparative Analysis on Edge Detection Techniques Used in Image
Processing”, IOSR Journal of Electronics and Communication Engineering (IOSRJECE), ISSN : 2278-
2834, vol. 1, issue 2, (2012), pp. 56-59.
[11] R. C. Gonzalez, R. E. Woods and S. L. Eddins, “Digital Image Processing Using MATLAB”, Pearson
Education Ptd. Ltd, Singapore, (2004).
[12] S. Jayaraman, S. Esakkirajan and T. Veerakumar, “Digital image processing”, Tata McGraw hill
education private limited, ISBN (13): 978-0-07-014479-8.
[13] P. Gupta, J.S. Kumare, U.P. Singh and R.K. Singh, ‘’Histogram Based Image Enhancement: A Survey”,
International Journal of Computer Sciences and Engineering, vol. 5, issue 6, (2017), pp.-177-182.
[14] Chauhan, A. Singh, S. Silakari and M. Dixit, “Image Segmentation Methods: A Survey
Approach”, 2014 Fourth International Conference on Communication Systems and Network
Technologies, (2014), pp. 929-933.
[15] M. Dixit and S. Silakari, “Hybrid Facial Feature Optimisation Approach Using Bezier Curve”, 2015
International Conference on Computational Intelligence and Communication Networks (CICN), (2015),
pp. 218 – 221.

10 Copyright ⓒ 2017 SERSC

 International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

Authors
Mohd. Aquib Ansari, he received the Bachelor of Engineering
degree in Information Technology from SATI, Vidisha, M.P. in 2014.
He is currently pursuing his Master of Technology from MITS,
Gwalior, M.P. He is vice chairperson of IEEE, MITS Student Chapter
as well as IEEE student member. His research interests include image
retrieval and image segmentation. He can be contacted at
[email protected].

Diksha Kurchaniya, she received the Bachelor of Engineering

degree in Computer Science and Engineering from ITM, Gwalior,
M.P. in 2014. She is currently pursuing her Master of Technology
from MITS, Gwalior, M.P. Her research interests include image
retrieval and image segmentation. She can be contacted at
[email protected].

Manish Dixit, he received the B.E. degree in the computer

technology from Barkatullah University, Bhopal and M.E. degree in
Communication Control and Networking from MITS, Gwalior in
1993 and 2006 respectively. He has completed his Phd. from Ragiv
Ghandhi Technical University, Bhopal in 2017. He is currently
working as an Associate Professor in the Department of Computer
Science & Information Technology, MITS, Gwalior, India. He has
presented various research papers in the National and International
Conferences and Journals. He is a Member of CSI, IETE, IAENG. He
is a member of IEEE & Secretary IEEE MP Sub-Section. He can be
contacted at [email protected].

Copyright ⓒ 2017 SERSC 11

 International Journal of Multimedia and Ubiquitous Engineering
Vol.12, No.11 (2017)

12 Copyright ⓒ 2017 SERSC

View publication stats