Circular discontinuities detection in welded joints using Circular Hough Transform
-
Upload
independent -
Category
Documents
-
view
3 -
download
0
Transcript of Circular discontinuities detection in welded joints using Circular Hough Transform
ARTICLE IN PRESS
0963-8695/$ - se
doi:10.1016/j.nd
�Correspondfax: +603 8925
E-mail addr
haniza.yazid@g
NDT&E International 40 (2007) 594–601
www.elsevier.com/locate/ndteint
Circular discontinuities detection in welded joints usingCircular Hough Transform
Hafizal Yazida,�, Haniza Yazidb, Mohd Haruna, Shukri Mohda, A. Aziz Mohameda,Mohamed Rizonb, Shaharudin Sayutia
aIndustrial Technology Division, Block 34, Malaysian Institute for Nuclear Technology Research (MINT), Bangi, 43000 Kajang, MalaysiabKolej Universiti Kejuruteraan Utara Malaysia (KUKUM), Jalan Kangar-Arau, 02600 Jejawi, Perlis, Malaysia
Received 16 November 2006; received in revised form 26 March 2007; accepted 25 May 2007
Available online 13 June 2007
Abstract
Conventional radiography is one of the common non-destructive testing which employs manual image interpretation. The
interpretation is very subjective and depends much on the inspector experience and working conditions. It is therefore useful to have
pattern recognition system in order to assist human interpreter in evaluating the quality of the radiographed sample, especially
radiographic image of welded joint. This paper describes a system to detect circular discontinuities that are present in the joints. The
system utilizes together two different algorithms, which are separability filter to identify the best object candidate and Circular Hough
Transform to detect the presence of circular shape. The result of the experiment shows a promising output in recognition of circular
discontinuities in a radiographic image. This is based on 81.82–100% of radiography film with successful circular detection by using
template movement of 10 pixels.
r 2007 Elsevier Ltd. All rights reserved.
Keywords: Non-destructive testing; Algorithm; Welded joints; Separability filter; Circular Hough Transform
1. Introduction
As one of the non-destructive testing (NDT), industrialradiography plays a vital role in detecting volumetricdiscontinuities present in the critical area of engineeringcomponent. The most common critical area is welded jointthat forms part of the structures. A variety of NDTmethods are available for the inspection of welded joint [1].The emerging trend of radiography as one of the NDTmethods is towards automated inspections [2–6]. This isfeasible as the information of the discontinuities is in theform of radiographic images. Interpretation and evaluationof the radiographic images are very important stages foracceptance of any engineering component. Realization ofthis crucial process dealing with images, image analysis and
e front matter r 2007 Elsevier Ltd. All rights reserved.
teint.2007.05.004
ing author. Tel.: +603 89250510x1194;
0907.
esses: [email protected] (H. Yazid),
mail.com (H. Yazid).
pattern recognition system using a computer have becomethe subject of interest in the field of industrial radiography.This trend of the radiographic interpretation and evalua-tion method is aimed towards an automated inspection,which results in better consistence assessment. Basically, itrequires development of image enhancement and objectdetection technique to assist human interpreter in evaluat-ing the quality of welded joints.The purpose of the radiographic interpretation is to
reach a decision on the acceptability or rejection of thecomponent being assessed. This is based on the adoptedcode and standard in use, for example ASME Code Section1 [7]. There are many types of discontinuities described bythe standard, which includes cracks, lack of weld penetra-tion, lack of weld fusion, porosity, voids and others.Every discontinuity has its own shape and contributes todifferent level of severity to the weldment. According toASME Code Section 1 certain types of discontinuities areclassified as major and unacceptable. These unacceptablediscontinuities are called defects and subject to repair and
ARTICLE IN PRESS
Contrast Enhancement
DefectSegmentation and Noise Reduction
Separability Filter
Good Image
Digitization Digitization
Manual FilmQualityInspection
Radiographic Image
Poor Image
H. Yazid et al. / NDT&E International 40 (2007) 594–601 595
reexamination. Such defects are cracks, lack of weldpenetration and lack of weld fusion. Other than thatincluding rounded discontinuity is still depending on thesize, distribution and concentration as specified by thestandard.
In this paper, the work is intensified in finding onetype of discontinuity that is circular and appeared asrounded image on the radiograph. The circular shape istreated as an object of detection. The approach is tooptimize separability filter technique and Circular HoughTransform (CHT) in order to obtain the result. The filtertechnique is used to recognize the best object candidatesand later identified by CHT. Similar approaches have beenused to detect coconut as circular object in coconut treeimage with prominent success rate by Haniza et al. [8].The use of CHT extends also to other research fieldssuch as iris detection for face recognition and fingertipsdetection [9,10].
CircularHough Transform
End result
Fig. 1. Flow procedure of the pattern recognition system.
1.1. Industrial radiography
In industrial radiography, typical images are generatedfrom using two types of radiation source, namely X-rayand gamma ray. Both differ primarily in energy andpenetration power. X-ray suffers lower penetration powerand energy as compared to gamma ray. The advantage ofX-ray over gamma ray is that it has a wider energyspectrum while energy for gamma ray is discrete [9]. Thisenables X-ray image to acquire higher latitude and bettercontrast. In other words, for most of the weldedcomponents, good image quality is attainable throughrecommended radiographic practice, thus, making theconventional radiography interpretation feasible to becarried out in an automated manner.
2. Methodology
A systematic procedure is established in analyzing theimage information and inferring patterns in such a waythat would yield the targeted object. The innovativeapproach in this procedure is to separate good imagequality and directly subject to separability filter and CHTwithout undergoing the preprocessing step. The otherimage still requires the preprocessing step for imageenhancement. Fig. 1 shows the flow procedure of thepattern recognition system. Initially the resulted radio-graphic film is subjected to film quality inspectionmanually. The inspection covers many criteria that includefilm density, sensitivity, presence of artifacts and identifica-tion. All the criteria have to be met as governed by theadopted standard in order for the radiographic film to beaccepted for evaluation. Even though the radiographic filmis accepted for evaluation, it does not mean that the imageappeared on the film is good in terms of contrast. A filmwould suffer poor radiography image quality or contrastonce the requirement of the criteria is only met on a
minimal basis. Therefore poor and good radiographyimage quality is identified before the film is digitized.
2.1. Preprocessing
Once poor radiography image quality is identified anddigitized, preprocessing step is initialized. The image issubjected to contrast enhancement. In this work, 8-bitimage is used as an input image which carries 0–255 graylevels. The image is in the form of PGM (Portable GrayMap) format and comprises 320� 240 pixels. Every singlepixel in an image carries its own intensity. The distributionof the intensities of the pixels could be viewed statisticallyby plotting the intensity histogram. By manipulating thishistogram, the required intensity distribution or contrast isachieved. It is common to have narrow intensity distribu-tion, which causes poor image contrast, and this requiresrectification to acquire a wider intensity distribution. Inthis work, histogram equalization technique has been usedto improve the image contrast. By using this technique, it ispossible to acquire a wider intensity distribution ascompared to the original intensity distribution profile.One typical example is clearly shown in Fig. 2. Thereforethe image contrast is effectively enhanced. This technique isparticularly suitable for region-based feature extraction.Typical radiography image which has good contrast and
does not require histogram equalization is depicted inFig. 3. Poor radiography image quality that requireshistogram equalization is depicted in Fig. 4. The resultof histogram equalization on poor radiography imagequality is depicted in Fig. 5. The same result of histogramequalization is depicted in Figs. 6 and 7, respectively.
ARTICLE IN PRESS
0
200
400
600
800
1000
1200
1400
0
11
22
33
44
55
66
77
88
99
11
0
12
1
13
2
14
3
15
4
16
5
17
6
18
7
19
8
20
9
22
0
23
1
24
2
25
3
Pix
el
Co
un
ts
Original Distribution After HistogramEqualization
Pixel Counts Vs.Gray Level
Gray Level
Fig. 2. The effect of histogram equalization on pixel counts distribution.
Fig. 3. Typical image, which has good contrast and does not require
histogram equalization.
Fig. 4. Typical image, which has poor contrast and requires histogram
equalization.
H. Yazid et al. / NDT&E International 40 (2007) 594–601596
2.2. Defect segmentation
This step is crucial to distinguish between the object ofinterest and the background. The term segmentationimplies segmenting the object of interest or foreground
from the background. Edge finding and mathematicalmorphology are used in this step. Edge-finding techniqueprovides detection of pixels that belong to the borders ofthe objects. As a result, the image contains only a collectionof edge pixels. The collection of edge pixels covers most ofour targeted object and also unwanted information ornoise. Unwanted information is reduced by using mathe-matical morphology. By doing this we will able to obtainedge pixel that mostly corresponds to the targetedobject. There are various edge-finding techniques availableand in this work Sobel edge detector has been used.This technique reveals thick edge as compared to othertechnique particularly Canny [10]. Sobel edge detector ispreferred due to its ability to reduce the chances of falseobject detection as compared to Canny. Canny reveals toomany details, which in turn cause the CHT to give a wrong-targeted object. Edge information is very important andcomes into play when performing CHT step. Sobel edgedetector is a gradient method, which detects the edges bylooking for the maximum and minimum in the firstderivative of the image. Gradient is reflected by thedifference in intensity from one pixel to the next. Sobeledge detector uses a pair of 3� 3 convolution masks, oneestimating the gradient in the X-direction, Gx and the otherestimating the gradient in the Y-direction, Gy. The Sobelconvolution masks are shown in Fig. 8.The result of each convolution mask is combined to get
exact magnitude value using the formula
jGj ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiGx2 þ Gy2
p. (1)
The Sobel technique deals with six pixels and thus has abetter average over the neighboring pixels. This approachwith slight blurring can eliminate most noise found in theimage.Mathematical morphology is considered as region-filling
technique to provide the complete objects. This technique
ARTICLE IN PRESS
Fig. 5. Image as in Fig. 4 after histogram equalization.
Fig. 6. Another image, which has poor contrast and requires histogram
equalization.
Fig. 7. Image as in Fig. 6 after histogram equalization.
Fig. 8. Sobel convolution masks.
H. Yazid et al. / NDT&E International 40 (2007) 594–601 597
is implemented by using closing and opening operation.First the closing operation is carried out followed byopening operation. The closing operation is able to fill insmall holes and removes much of the white pixel noise.This would yield a fairly clean image as compared toprevious image before the operation. Then opening opera-tion takes place by opening small gaps or spaces betweentouching objects in an image. The opening operation canseparate objects that are connected in a binary image. Bothoperations generate a certain amount of smoothing on anobject contour given a ‘‘smooth’’ structuring element.
2.3. Separability filter
After completing segmentation step, the image issubjected to feature extraction step by using separabilityfilter technique. This technique is a template-based methodand was proposed by Fukui [11]. This technique employs atemplate of Fig. 9 by placing the template at each point(xi, yi) and then computation on the separability betweenthe two regions R1 and R2 in the template with size r iscarried out.Eq. (4) is used to measure the separability between the
two regions. r is treated as a fixed parameter based onmaximum acceptable circular defect size specified by thestandard. r is fixed at 3, 4 and 5 pixels that correspondapproximately from 0.75 to 1.5mm defect radius.
A ¼XN
i¼1
ðIðxi; yiÞ � PmÞ2, (2)
B ¼ n1ðP1 � PmÞ2þ n2ðP2 � PmÞ
2, (3)
Z ¼B
A, (4)
where nk (k ¼ 1,2) is the number of pixels in Rk;
N ¼ n1 þ n2; Pk (k ¼ 1,2) the average intensity in Rk; Pm
the average intensity in the union of R1 and R2; and Iðxi; yiÞ
the intensity values of pixels ðxi; yiÞ in the union of R1
and R2.
Fig. 9. Circle template for object candidate detection.
ARTICLE IN PRESS
Fig. 11. Image space.
Fig. 12. Parameter space.
H. Yazid et al. / NDT&E International 40 (2007) 594–601598
The points (xi,yi) obtained after the opening morphologyin the segmentation process is used as the input to theseparability filter. The range of diameter for the R1 is set to{2rL,y,2rU} and in this work rL is 3 pixels and rU is 5pixels. Diameter of R2 region is obtained after consideringthe gap of 0.5r. The template of Fig. 9 will move every x
pixels from center to the left and right and y pixels fromcenter to upwards and downwards as shown in Fig. 10. Thecircles that give the local maxima of the separability Z areselected as object candidates.
In this work; x ¼ 10 pixels, y ¼ 10 pixels.
2.4. Circular Hough Transform
Circular Hough Transform or CHT is used as patternclassification to identify the targeted object. This is the laststep need to be carried out before obtaining the end result.Hough transform is a widespread technique in imageanalysis. It was first introduced by Paul Hough in 1962. Itwas first used to detect straight lines and later extended tocater to other parametric shapes such as circular, ellipsesand finally being generalized to cater to any parametricshape. CHT was proposed by Duda et al. [12]. It has beenrecognized as a robust technique for curve detection.
CHT is used to transform a set of feature points in theimage space into a set of accumulated votes in a parameterspace. Therefore the difficulties in global detection problemin the image space are reduced to peak finding in aparameter space. The image space and parameter space areshown in Figs. 11 and 12, respectively. In this work, theinteresting image points are those that belong to a circlecharacterized by its center coordinates (x0, y0) and itsradius r can be represented by
ðx� x0Þ2þ ðy� y0Þ
2¼ r2. (5)
This equation indicates that a point (x, y) of the imagespace can be mapped to a cone surface in the (x0, y0, r)parameter space, which represents all possible circlespassing through (x, y). In this work, the edge orientation
Fig. 10. Template movement.
information is used to reduce the ranges of parameters andthe complexity of the voting operation. The center of acircle must be on the straight line passing through the edgepoint (x, y) in a direction given by the gradient value at theedge point. The use of edge orientation information isproposed by Kimme et al. [13] and this limits the possiblepositions of the center for each edge point. This methodapplies the use of an arc plotted perpendicular to the edgeorientation at a distance r from the edge point.Each edge point contributes a circle of radius r to an
output accumulator or parameter space. The outputaccumulator space has a peak where these contributedcircles overlap at the center of the original circle. The peakpoint is projected back to the image space as (x0, y0, r)which reveal the final detected circle object. This circleconsists of many edge points at its perimeter.Hough transform and its modified version offer several
advantages. First the possibility of parallel processing sinceeach image point is treated independently and thealgorithm is able to work even when the shape is occluded.Secondly, the method is robust even in the presence ofnoise. Noisy image points are very improbable tocontribute to a peak in the parameter space. Thirdly,the method is able to detect different instances of thedesired shape at the same time, depending on the numberof peaks, which are considered in the parameter space.
ARTICLE IN PRESSH. Yazid et al. / NDT&E International 40 (2007) 594–601 599
The advantages over other techniques make it an attractivemethod for circle detection in many research fields.
Fig. 13. Typical image after applying Canny edge detector for image in
Fig. 3.
Fig. 14. Typical image after applying Sobel edge detector for image in
Fig. 3.
3. Results and discussions
The system was tested by evaluating 11 radiographyimages. The radius of the object as the input parameter wasset to 3, 4 and 5 pixels, which correspond to approximately1.5–3.0mm defect diameter size. Table 1 shows the resultsof the experiment.
First and second detections successfully detect thepresent of circular discontinuities for all the radiographyimages. Typical successful detection is depicted in Fig. 17.The same image before the detection is depicted inFig. 16. Third and fourth detections also give higherpercentage of successful detection with 90.9% and 81.82%,respectively.
The reduction in successful or correct detection is due tothe present of ambiguous object detected by the system asin Fig. 19. Image before detection is depicted in Fig. 18. Itis not surprising for CHT to give false detection since thepresence of ambiguous object does exist after applyingSobel edge detector as in Fig. 14. Thus the performanceof the proposed system is far better if the ambiguousobject outside the region of interest is eliminated afterapplying the edge detector. Fifth detection gives 54.55% ofsuccessful detection. Almost half out of the total film isable to give correct detection. Besides the factor ofambiguous object, this is also an indicator that all thecircular discontinuities that are present in the area ofinterest have been detected by the system and the rest of thedetection become false detection in order for a computerprogram to complete.
In this work, Sobel edge detector has been selected.This type of edge detector would reduce the amount ofambiguous object and reveal cleaner image as compared toCanny edge detector. Both of the edge detectors aredepicted in Figs. 13 and 14 for comparison. Some of theimage does not require histogram equalization due to thefact that it has already a good image quality and contrast.This typical image is depicted in Fig. 3. On the contrary,poor image quality and contrast image are requiredto undergo histogram equalization. This typical imageis depicted in Fig. 4. The same image after applyinghistogram equalization is depicted in Fig. 5. The improved
Table 1
Success in circular detection
No. of success in circular
detection
Percentage of film according to
success no. of detection (%)
1 100
2 100
3 90.9
4 81.82
5 54.55
Fig. 15. Effect of mathematical morphology. Image (a) indicates image
before being subjected to mathematical morphology. Image (b) indicates
image after being subjected to mathematical morphology.
image contrast after histogram equalization eases into thefollowing process flow of object detection. Additionally, inthis work, the effect of mathematical morphology eases theburden of CHT by removing pixels or unwanted informa-tion and thus leaves only potential pixels for CHT to do the
ARTICLE IN PRESS
Fig. 17. Image as in Fig. 16, after detection.
Fig. 16. Image before detection.
Fig. 18. Image before detection.
Fig. 19. Image as in Fig. 18, after detection.
H. Yazid et al. / NDT&E International 40 (2007) 594–601600
analysis. This is depicted in Figs. 15–19. Every single stepin the process flow is essential especially before theapplication of separability filter and CHT technique.
4. Conclusions
The work could be considered as an evidence of theapplicability of the separability filter and CHT in thedetection of circular object in the radiography image.A 100% of film was found to have first and secondcorrect circular detections by the proposed system.Third and fourth detections also give higher film percen-tage of successful detection with 90.9% and 81.82%,respectively. This indicates that the system is suitablefor the use of circular detection in the welding area.By having a proper image quality and contrast coupledwith Sobel edge detector, the circular object is easilydetected by separability filter and CHT technique. Addi-tionally, this work shows good image contrast orquality could avoid image enhancement in the preproces-sing step of the pattern recognition flow procedure. Eventhough the proposed system is efficient in circulardetection, care has to be taken in the edge detector stageto avoid ambiguous object that could lead to falsedetections. Further research will be focused on finding away to avoid the presence of ambiguous object and findinga technique to accurately determine the size of the detectedcircular object.
Acknowledgments
The authors wish to thank MINT’S NDT group andKhair’iah Yazid for providing the image data andSheriffah Noor Khamseah for proofreading the paper.The excellent support from MINT and KUKUM through-out the research is highly appreciated.
References
[1] Edward GR. Inspection of welded joints. ASM handbook, welding,
brazing and soldering, vol. 6. Materials Park, OH: ASM Interna-
tional; 1993. p. 1081–8.
[2] Boerner H, Strecker H. Automated X-ray inspection of aluminium
castings. IEEE Trans Pattern Anal Mach Intell 1988;PAMI-10(1):
79–91.
[3] Demandt K, Hansen LK. Real-time X-ray system with fully
automated defect detection and quality classification. In: Halmshaw
R, editor. X-ray real time radiography and image processing,
Proceedings of symposium, Newbury, Berkshire; 1988. p. 96–119.
[4] Gayer A, Sayer A, Shiloh A. Automatic recognition of welding
defects in real time radiography. NDT Int 1990;23(3):131–6.
[5] Builtjes JH, Rose P, Daum W. Automatic evaluation of weld
radiographs by digital image processing. In: Halmshaw R, editor.
X-ray real time radiography and image processing, Proceedings of
symposium, Newbury, Berkshire; 1988. p. 63–72.
[6] Munro JJ, et al. Weld inspection by real time radioscopy. Mater Eval
1987;45(11):1303–9.
[7] American Society of Mechanical Engineers (ASME), Section 1; 1998,
p. 204.
ARTICLE IN PRESSH. Yazid et al. / NDT&E International 40 (2007) 594–601 601
[8] Yazid H, et al. Object detection using circular Hough transform. Am
J Appl Sci 2005;2(12):1606–9.
[9] D’Orazio T, Guaragnella C, Leo M, Distante A. A new algorithm for
ball recognition using circle Hough transform and neural classifier.
Pattern Recognition 2004;37:393–408.
[10] Kawaguchi T, Hidaka D, Rizon M. In: Proceedings of IEEE ICRP,
vol. 1; 2000, p. 49–52.
[11] Fukui K, Yamaguchi O. Facial feature points extraction method
based on combination of shape extraction and pattern matching.
Trans IEICE Jpn J80-D-II 1997;8:2170–7.
[12] Duda RO, Hart PE. Use of the Hough transformation to detect lines
and curves in picture. Commun ACM 1972;11–5.
[13] Kimme C, Ballard D, Sklansky J. Finding circles by an array of
accumulators. Commun ACM 1975;18(2):120–2.