A study on pre-processing algorithms for metal parts inspection

Abstract

Pre-processing is very useful in a variety of situations since it helps to suppress information that is not related to the exact image processing or analysis task. Mathematical morphology is used for analysis, understanding and image processing. It is an influential method in the geometric morphological analysis and image understanding. It has befallen a new theory in the digital image processing domain. Edges detection and noise reduction are a crucial and very important pre-processing step. The classical edge detection methods and filtering are less accurate in detecting complex edge and filtering various types of noise. This paper proposed some useful mathematic morphological techniques to detect edge and to filter noise in metal parts image. The experimental result showed that the proposed algorithm helps to increase accuracy of metal parts inspection system.

Full text

Page 1

A Study on Pre-processing Algorithms for
Metal Parts Inspection

Haider Sh. Hashim, Anton Satria Prabuwono, Siti Norul Huda Sheikh Abdullah
Center for Artificial Intelligence Technology (CAIT),
Faculty of Information Science and Technology,
University Kebangsaan Malaysia, 43600 UKM Bangi, Selangor D.E., Malaysia
haider_hashim76@yahoo.com; antonsatria, mimi{@ftsm.ukm.my}


Abstract— Pre-processing is very useful in a variety of situations
since it helps to suppress information that is not related to the
exact image processing or analysis task. Mathematical
morphology is used for analysis, understanding and image
processing. It is an influential method in the geometric
morphological analysis and image understanding. It has befallen
a new theory in the digital image processing domain. Edges
detection and noise reduction are a crucial and very important
pre-processing step. The classical edge detection methods and
filtering are less accurate in detecting complex edge and filtering
various types of noise. This paper proposed some useful
mathematic morphological techniques to detect edge and to
filter noise in metal parts image. The experimental result
showed that the proposed algorithm helps to increase accuracy
of metal parts inspection system.

I.
INTRODUCTION
Pre-processing is very useful in a variety of situations
since it helps to suppress information that is not related to the
exact image processing or analysis task. Image filtering and
edge detection is an indispensable and necessary operation in
image pre-processing step [1]. When the image is acquired,
the factors such as the projection, mix, aberrance and noise
are produced. These factors can lead to image blurring and
distortion; consequently it can cause difficulty in extracting
important feature in the image [2] like edges of objects in
image. Mathematical morphology is a mathematical method
using structural elements to analyze image, which has been
widely applied to variety fields [3] like visual inspection. The
morphological operations work to simplify image and
preserve the main shape characteristics of objects.
Morphological operators are mainly used in image pre-
processing for (noise filter, shape simplification and
enhancing object structure) [4]. It provides an option
approach to image processing depending on a shape set
theory [5]. Mathematical morphology is a powerful tool to
filter different types of noise and to detect the complex edge
in variety directions [1, 2]. Edge detection is one of the basal
contents in the image processing and analysis, which is
unable to be resolved completely so far [6]. This paper
objective is to evaluate pre-processing algorithms for metal
parts inspection system. We organize this paper into five
sections. The second and third sections describe the basic
mathematical morphology theory and proposed methodology
subsequently. Then, the forth section explains the
experimental evaluations. Finally, the last section wraps up
the paper.
II.
BASIC MATHEMATICAL MORPHOLOGY THEORY
Morphology operators like dilation and erosion with
structuring elements (SE) are the most fundamental operators
in mathematical morphology and have become common tools
for both image filtering and analysis [3]. In 1960s, Matheron
introduced mathematical morphology as a method for
analyzing a geometric structure of metallic and geologic
samples. Later, Serra is developed the method for image
analysis. Mathematical morphology is a very significant
theory, whose operation must be defined by using arithmetic
sets [2]. Consequently, the resultant image posses two sets of
different criteria such as determining either black and white
pixel at specific location
morphology like addition, multiplication or etc. It is an
influential tool for dealing with a variety of problems in
image processing and machine vision [3]. Basically,
mathematical morphology
morphological algebraic arithmetic operators. The essential
morphological operations are dilation, erosion, closing,
opening. Those operators are used for detecting, modifying,
manipulating features in a predetermined image depending on
their shapes. In line with that, morphological operators are
closely related to the size and shape of structuring element
(SE). SE depends on the purpose and requirements of
application connected to the image [4]. We preview the basic
of mathematical morphological operators of gray-scale
images. Suppose
is an input image,
SE, and
are the domains of the input image and the SE,
respectively:
Definition 1: Dilation operation
based on mathematical
consists of a series of
and is a
, defined as,

Definition 2: Erosion operation
, defined as,

Definition 3: Opening operation
, defined as,

Proc. 1st International Conf. Energy, Power and ControlBasrah University, Basrah, Iraq 30 Nov. to 2 Dec. 2010
1
Iraq J. Electrical and Electronic Engineering
Vol.7 No.1, 2011
ﺔﻴﻧﻭﺮﺘﻜﻟﻻﺍﻭ ﺔﻴﺋﺎﺑﺮﻬﻜﻟﺍ ﺔﺳﺪﻨﻬﻠﻟ ﺔﻴﻗﺍﺮﻌﻟﺍ ﺔﻠﺠﻤﻟﺍ
2011 ,1ﺩﺪﻌﻟﺍ ,7ﺪﻠﺠﻣ
________________________________________________________________________________________________________________________

Page 2

Definition 4: Closing operation
, defined as,
)

From the definition 1 and 2, erosion is a change of
shrinking, which decreases the specific pixel detail in an
image, while dilation is a change of expanding, which
increases the specific pixel detail in an image. However, both
methods are sensitive to the image edge whose value changes
apparently. Erosion filters the inner boundary of region of
interest (ROI) in an image while dilation filters the outer ROI
image. From the definition 3 and 4, opening is erosion
followed by dilation, while closing is dilation followed by
erosion. The opening operation is employed for removing
small bright details (peak noise) smaller than the structural
elements, smoothing the boundary of a ROI image, breaking
narrow gaps, whilst the closing operation is applied for
eliminating the small dark details (low noise), smaller than
the structural elements, fusing narrow breaks or gaps,
eradicating small holes and filling up gaps in the contours of
ROI image [1]. Consequently, combining these two
operations for image filtering can effectively remove peak
and reduce noise in an image. As conclusion, morphological
operations are usually used to detect image edge and to
remove unwanted information in the image.
III.
PROPOSED FRAMEWORK
Fig. 1 shows the proposed framework for metal parts
visual inspection. The pre-processing steps are necessary for
object recognition, noise reduction and edge detection. An
image enhancement is aimed to provide a better image
compared to the source image. We remove image noise and
detect edge of objects by using mathematical morphology
with SE. In conjunction to metal parts inspection, a post-
processing algorithm evaluation is conducted to classify the
right metal parts in real-time inspection.






















Figure 1. Proposed framework for metal parts visual inspection.
A. Pre-processing
Pre-processing is one of important steps in image
processing. We apply some useful basic morphological
operations such as opening and closing operators.
1) Image Enhancement
The morphological SE filter has a strong adaptability. It
has good filtering effects on three different kinds of noise.
Besides, it is powerful to filter out peak and low noise (salt
and pepper). It can also maintain the detailed information of
image. The most basic morphological operations are dilation
and erosion. We apply some useful basic morphological
operations such as Opening and Closing Operators.
2) Edge Detection
An edge is a sharply change of intensity in the image. The
techniques can be classified based on first order derivative
and second order derivative. The first technique uses gradient
vector and estimates the gradient direction, e.g. Roberts,
Sobel, and Prewitt. Then, the second technique uses the zero-
crossing of the Laplacian or non-linear differential [7]. We
apply the Morphological filter and edge detection based on
SE. Generally, it is show a stronger method to filtering effect
for images with peak noise (such as speckle noise), and it can
work with effectively preserve the details of the images and
detects the important edges.
Based on experimental results, we could only achieve
satisfactory results for real images after using diamond SE
with
window size as shown in Fig. 2(a). We choose
for the
direction angles as illustrated in Fig. 2(b) [8].










B. Post-processing
In the post processing, we use fuzzy logic to classify the
metal part. Fuzzy logic provides a natural method for
constructing systems that emulate human decision making
processes. Fuzzy ruled-based systems are knowledge based
systems in which the knowledge is stored in a knowledge
base that consists of three components: linguistic variables,
membership functions, and a set of “IF-THEN” rules. Fuzzy
inference systems have been successfully applied in fields
such as automatic control, data classification, decision
analysis, expert systems, and computer vision. There are
mainly two types of fuzzy inference systems: 1) the Mamdani
type and 2) the Sugeno type. The main difference between the
Mamdani and Sugeno types is that the Sugeno output
membership functions are either linear or constant. Based on
our experimental results, we obtained an optimized model of
this problem using Mamdani fuzzy inference rule [9].
•
•
1
•
•
Fig.2 (a). Simple of SE filters
with
•
1
1
1
•
1
1
1
1
1
•
1
1
1
•
•
•
1
•
•
.
•
1
•
•
•
Fig.2 (b). Simple of SE edge
detection with
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
1
•
; angle of 145o.
Image acquisition
Using
Web camera
Light source
Conveyor
belt
Image enhancement
Using
Morphology
Filtering SE
Edge detection
Using
Morphology
SE
Features extraction
Using
Morphology
SE
Classification
Using
Fuzzy logic
2
Iraq J. Electrical and Electronic Engineering
Vol.7 No.1, 2011
ﺔﻴﻧﻭﺮﺘﻜﻟﻻﺍﻭ ﺔﻴﺋﺎﺑﺮﻬﻜﻟﺍ ﺔﺳﺪﻨﻬﻠﻟ ﺔﻴﻗﺍﺮﻌﻟﺍ ﺔﻠﺠﻤﻟﺍ
2011 ,1ﺩﺪﻌﻟﺍ ,7ﺪﻠﺠﻣ
________________________________________________________________________________________________________________________

Page 3

IV.
EXPERIMENTAL RESULT
In this section, we present the experiment results. The
proposed algorithm of edge detection and image filtering
used mathematic morphological SE and compare with a
different of traditional edge detection methods. Fig. 3(a)
shows the original image that captured using web cam with
size of
. We tested on 10 images where the total
number of actual objects is about 100 ROI (10 objects 10
images). We then classified those ROI using Mamdani fuzzy
logic into 10 classes such as screw, big shim, small shim,
medium shim, very small shim, slim bolt, short thin bolt, long
thin bolt, big nut, and medium nut. To evaluate the overall
performances, we calculate the accuracy as the followings:
??????????????????????????? =

Fig. 4(a) and Fig. 5(a) show an original image with salt
and pepper noise. Furthermore, Fig. 3(b), 3(c), and 3(d) are
the results after applying Gaussian filter, Sobel operator and
after classifying using fuzzy rule respectively. Fig. 4(b), 4(c),
and 4(d) are the results after applying Average filter, Canny
operator and after classifying using fuzzy rule subsequently.
Finally Fig. 5(b), 5(c), and 5(d) are the results using the
proposed algorithm.
From Table I, Sobel with Gaussian filter, Canny with
Average filter and proposed framework obtained about 55%,
63% and 97% respectively. As conclusion, we can conclude
that the proposed framework can detect edge and filter noise
better than other conventional methods.

TABLE I.
THE ACCURACY OF CLASSIFICATION









V.
CONCLUSION
In this paper, mathematical morphological operators are
used for noise filtering whilst the edge detection using
boundary tracing algorithm is proposed. The experimental
result showed the proposed algorithm is more efficient than
the conventional operator used such as Sobel with Gaussian
and Canny with Average. The Morphological SE result is
clear and continual edges. It provides all important
information of object under inspection process. Sobel
operator could be applied only for two directions e.g. vertical
and horizontal. It use small mask to scan the whole image
namely time consuming and did not detect the edge with any
angle directions. Canny can detect a lot of information of
image. It is not suitable to apply in machine vision since it
needs to filter the information for the next process. The
proposed algorithm could detect the edge in different and
complex direction and show clear edge, pinpointed, integral
and continual results. The final results showed the accuracy
of proposed framework was 97%.
ACKNOWLEDGMENT
Special thanks to Dr. Adnan Jabir Mr. Hassan Majeed
who had worked hard in giving ideas and to effort in various
areas. This project research is funded by Research University
Operation No. OUP-ICT-36-186/2010, Junior Researcher
Grant No. UKM-GGPM-ICT-103-2010 and UKM-AP-ICT-
17-2009.
REFERENCES

[1] Y. Zhao, W. Gui, and Z. Chen, “Edge detection based on multi-
structure elements morphology,” Proc. the 6th World Congrees on
Intelligent Control and Automation, pp. 9795-9768, Dalian, October
2006.
[2] F.Y. Cui, L.J. Zou, and B. Song, “ Edge feature extraction based on
digital image processing techniques,” Proc. IEEE International
Conference on Automation and Logistics, pp. 2320-2324 , Qingdao , 30
September 2008.
[3] G.B. Xu, Z.B. Su, J. Wang, Y.X. Yin, and Y.L. Shen, ”An adaptive
morphological filter based on multiple structure and multi-scale
elements,” Proc. Second International Symposium on Intelligent
Information Technology Application, Vol. 2, pp. 399-403, Shanghai, 6
January 2008.
[4] Y. Tingfang, L. Pei, Z. Xiangjun, and K.K. Li, “Application of adaptive
generalized morphological filter in disturbance identification for power
system signatures,” Proc. International Conference on Power System
Technology, pp. 1-7, Chongqing, 2006.
[5] Y.T. Hsiao, C.L. Chauang, J.A. Jiang, and C.C. Chien, ”A contour
based image segmentation algorithm using morphological edge
detection,” Proc. IEEE Internaional Conference on Systems Man and
Cybernetics, Vol. 3, pp. 2962-2967, 2005.
[6] H. Akbar, and A.S. Prabuwono, ”Webcam based system for press part
industrial inspection,” International Journal of Computer Science and
Network Security, Vol. 8, No. 10, pp. 170-177, October 2008.
[7] M.I. Rajab, M.S. Woolfson, and S.P. Morgan, “Application of region-
based segmentation and neural network edge detection to skin lesions,”
Computerized Medical Imaging and Graphics, Vol. 28, No. 1-2, pp. 61-
68, January 2004.
[8] E.R. Urbach, and M.H.F. Wilkinson, ”Efficient 2-D grayscale
morphological transformations with arbitrary flat structuring elements,”
IEEE Transaction on Image Processing, Vol. 17, No. 1, pp.1-8, Jan
2008.
[9] H.Sh. Hashim, S.N.H.S. Abdullah, and A.S. Prabuwono, ”Automated
visual inspection for metal parts based on morphology and fuzzy rules,”
Proc. International Conference on Computer Application & Industrial
Electronics, Kuala Lumpur, 2010.















Sobel
& Gaussian
Canny
& Average
Proposed
method
Total no. of object 100 100 100

Classified accuracy


55% 63% 97%
3
Iraq J. Electrical and Electronic Engineering
Vol.7 No.1, 2011
ﺔﻴﻧﻭﺮﺘﻜﻟﻻﺍﻭ ﺔﻴﺋﺎﺑﺮﻬﻜﻟﺍ ﺔﺳﺪﻨﻬﻠﻟ ﺔﻴﻗﺍﺮﻌﻟﺍ ﺔﻠﺠﻤﻟﺍ
2011 ,1ﺩﺪﻌﻟﺍ ,7ﺪﻠﺠﻣ
________________________________________________________________________________________________________________________

Page 4

(a) (b) (c) (d)
Figure 4. The result of experiment using Average and Canny: (a) Image with Salt and Pepper noise, (b) Image filtered by
Average filter, (c) The results of edge detection used Canny operator, (d) Classification uses Fuzzy Rules.



(a) (b) (c) (d)
Figure 5. The result of experiment using proposed algorithm: (a) Image with Salt and Pepper noise, (b) Image filtered by Structure
elements filter, (c ) The results of edge detection using structure elements morphology, (d ) Classification uses Fuzzy Rules.





(a) (b) (c) (d)
Figure 3. The result of experiment using Gaussian and Sobel: (a) Original Image size
filter, (c) The results of edge detection using Sobel operator, (d) Classification uses Fuzzy Rules.
, (b) Image filtered by Gaussian










4
Iraq J. Electrical and Electronic Engineering
Vol.7 No.1, 2011
ﺔﻴﻧﻭﺮﺘﻜﻟﻻﺍﻭ ﺔﻴﺋﺎﺑﺮﻬﻜﻟﺍ ﺔﺳﺪﻨﻬﻠﻟ ﺔﻴﻗﺍﺮﻌﻟﺍ ﺔﻠﺠﻤﻟﺍ
2011 ,1ﺩﺪﻌﻟﺍ ,7ﺪﻠﺠﻣ
________________________________________________________________________________________________________________________