Edge detection in an infrared image in steel and metallurgical domain
K. Gherfi
a,b, M. Tria
a, H. Bendjama
a, R. Boulkroune
a, D. Idiou
a, L. Cherrad
aa
Research Center in Industrial Technologies (CRTI), BP 64 Cheraga,
Algiers 16000, Algeria
b
Laboratoire Automatique et Signaux Annaba Université Badji Mokhtar, BP 12 Annaba 23000,
Algeria
k.gherfi@crti.dz, kaddour.gherfi@gmail.com
T. Bensouci
c,dc
Laboratoire de T. du Signal, Département d’Electronique
Université Constantine 1, Route Ain El-bey, Constantine 25011, Algeria
d
Département de Télécommunication, Faculté d’Electronique et Informatique USTHB, BP. 32, Bab-Ezzouar, Algeria
Abstract—The iron and steel domain has a very important role in the industry. Indeed, much research has been fixed to the product quality, which is an essential factor in the production cycle. Among the most used techniques to today, to inspect the finished product, there is the infrared thermography. This latter allows for detecting defects in materials. Treatments of captured image using an infrared camera can detect the edge of these defects and hence locate them. In this work we try to apply mathematical methods to detect these edges.
Keywords— Infrared thermography; materials; defects; edges.
I. INTRODUCTION
Recently, investigation and detection of default in systems and materials has an important role in processing image.
The edge detection technique is usually used to extract of edge characteristics of the image. Edge characteristics are one of the most principal and essential characteristics of image and they can be use to determine the boundaries of the image. In recent years, many new methods are proposed in image processing domain.
Many edge detection methods are proposed unceasingly in past decades [1–3]. Among the edge detection methods, a gradient edge detection method is one of the most important methods to detect the edge of the objects in image processing. In this method there are first order derivative and second order derivative.
These methods used the vertical, horizontal or diagonal gradients to adjust the direction of the edge [4–7],
There are also several edge detection algorithms are presented in literature such as Canny, Sobel, Robert and Prewit [8-12]. These operators are capable to detecting the edge fast but the precision is depends of the operator.
In this paper, different methods of the edge detection are presented, to know the difference between operators in the case of infrared image.
II. DERIVATES METHODS
The edge concept being interconnected to the variation, it is clear that such definition we naturally lead to an evaluation of the variation in each pixel. It exists a variation if the gradient is locally maximum or if the second derivative (to be defined in a two-dimensional space) has a zero crossing.
The main known algorithms (Sobel, Prewitt, Kirsch, Canny, Deriche ...) focus on this first aspect of the edge detection method [13].
The derivates methods rely on the fact that the image edges are generally translated by the quick transitions of the image, and the slow variations will be eliminated by derivation. The edge of objects in any size images (natural images 2D, medical images 2D and 3D) usually correspond to local extreme values of the gradient or zero Laplacian of the grayscale function.
A. First Order Derivative (Gradient method)
In this method we detect the edges by looking for the maximum and minimum of the grayscale of image.
This technique is the most influential in the image processing, if we have a function f(x,y), the magnitude of the gradient of f at coordinates (x,y) is defined as :
h
j i f j h i y f x
x
f
) , ( ) , ) (
,
( + −
=
∂
h
j i f h j i y f x
y
f
) , ( ) , ) ( ,
( + −
=
∂
( ( , ) )
2( ( , ) )
2) ,
( x y f x y f x y
f = ∂
x+ ∂
yB. Second Order Derivative (Gradient method)
In this case we use the second derivative in the place of first derivative:
) , ( ) , ( 2 ) , ) (
,
2
(
j h i f j i f j h i x f
y x
f = + − + −
∂
∂
) , ( ) , ( 2 ) , ) ( ,
2
(
h j i f j i f h j i y f
y x
f = + − + −
∂
∂
2 2 2 2
) , ( )
, ) (
,
(
∂ + ∂
∂
= ∂
y y x f x
y x y f
x f
C. Roberts method
Roberts’s method is a discrete approach of the derivative with a step equal to 1, this function is recognized by the gradient of a function.
The Roberts operator described in a rectangular image having a ramp edge, it is calculated in 4 points, this amounts to convolve the image with both filters Rx=[-1 1] and Ry=transpose ([-1 1 ]).
D. Prewitt method
It is a derivative operator, it is calculated at 9 points, this filter performs a local average in 3 points. It is defined by a double mask. The matrix corresponding to the horizontal filtering essentially making out the vertical edges.
E. Sobel method
This filter is a derivative operator, it is calculated at 9 points, in the same manner as the Prewitt filter, but makes it possible to favor the calculation according to certain directions (horizontal, vertical, oblique).
In this case we get a better result by replacing the rectangular filter with a triangular filter.
F. Canny method
This filter eliminates false edges. Considering not only the intensity of the gradient but also its direction, it is possible to eliminate a pixel that points to two pixels with higher value because it is not a local maximum.
III. EXPERIMENTAL RESULTS
This study presents two methods of image edge detection based on gradient, first order derivative and second order derivative. These edge detection methods applied on steel sample with defects, many operations were investigated of this image, we present the original image and their edges detection in below figures, with the first and second derivative gradient methods and with some methods as Sobel, Prewitt and Canny detectors.
In these edges we focus on the limits of defects in image, and we try to see the difference between each algorithm in edge detection and the best method.
Figure 1: Original image
Figure 2: First Order derivative edge image
Figure 3: second Order derivative edge image
Figure 4: Roberts edge image
Figure 5: Prewitt edge image
Figure 6: Sobel edge image
Figure 7: Canny edge image
From these results we see that there is a big difference between the first order derivative method and the second order derivative method, this difference notes in the appearance the noise outside of the default and the default appears without precision.
We know that Roberts, Sobel, Prewitt and Canny based on the first order derivative (gradient), but we see that Prewitt and Sobel methods produce almost the same edge map, this edge map is better than the Roberts method.
Canny method produces an edge map better than Sobel, Prewitt and Roberts but with some noise.
IV. CONCLUSION
This study presented an application of several methods of edge detection on an infrared image of steel material with default.
We present the results of the first and second derivative gradient methods, Robert, Prewitt, Sobel and Canny.
From these results we conclude that, if we want to present edges of defects with precision we use canny method, because the edge map is the better and less sensitive to noise, in the other hand, if we want to present edges of defects without noise outside of it and with noise inside of it, we use the second order derivative method.
However, the second order derivative method and the Canny edge detection method produce the better edge map.
V. REFERENCES
[1] P. Melin, O. Mendoza, O. Castillo, An improved method for edge detection based on interval type-2 fuzzy logic, Expert Syst. Appl. 37 (12) (2010) 8527–8535.
[2] D.O. Aborisade, Novel fuzzy logic based edge detection technique, Int. J. Adv. Sci. Technol. 29 (2011) 75–82.
[3] S. Isik, K. Ozkan, A novel multi-scale and multi-expert edge detection method based on common vector approach, in: Signal Processing and Communications Applications Conference (SIU), 2014 22nd, IEEE, 2014, pp. 1630–1633.
[4] Lee Jonghwa, Jeong Taeu, Lee Chuhee. Edge-adaptive demosaicking for artifact suppression along line edges.
IEEE Trans Consum Electron 2007;53(3):1076–83.
[5] Tsai ChiYi, Song KaiTai. Heterogeneity-projection hard- decision color interpolation using spectral-spatial correlation. IEEE Trans Image Process 2007;16 (1):78–91.
[6] Jimmy Li Jim S, Randhawa Sharmil. Color filter array demosaicking using high-order interpolation techniques with a weighted median filter for sharp color edge preservation.
IEEE Trans Image Process 2009;18(9):1946–57.
[7] Yun Se-Hwan, Kim Jin Heon, Kim Suki. Color interpolation by expanding a gradient method. IEEE Trans Consum Electron 2008;54(4):1531–9.
[8] A. Rosenfel, Computer vision: a source of models for biological visual process,IEEE Trans. Biomed. Eng. 36 (1) (1989) 83–94.
[9] I. Sobel, Neighbourhood coding of binary images fast contour followingand general array binary processing, Comp. Graph. Image Process. 8 (1978)127–135.
[10] D. Marr, E.C. Hildreth, Theory of edge detection, Proc.
R. Soc. 207 (1167) (1980)187–217.
[11] J. Canny, A computational approach to edge detection, IEEE Trans. Pattern Anal.Mach. Intell. 8 (6) (1986) 679–
698.
[12] C.G. Rafael, E.W. Richard, L.E. Steven, Digital Image Processing Using MATLAB, Publishing House of Electronics Industry, Beijing, 2005.
[13] Muthukrishnan.R and M.Radha, edge detection techniques for image segmentation, International Journal of Computer Science & Information Technology (IJCSIT) Vol 3, No 6, Dec 2011.