AUTOMATIC DEFECT DETECTION AND
CLASSIFICATION OF CERAMIC TILES
Patil A.G.1, Phakade S.V.2, Shirsath D.O.3
1,2,3
Department of E&Tc, P.V.P.I.T. Budhgaon,(India)
ABSTRACT
This paper presents the development and application of image analysis and computer vision system in defect
detection of ceramic tiles. Computer vision is a rapid, consistent and objective inspection technique, which has
expanded to varied industries. Monitoring and detecting defect is becoming a very important issue in ceramic
tiles industry since maintaining rate of production with respect to time.
For the ceramic tile industry maintaining quality is more important, bulk amount of ceramic tiles are
manufactured; it is very difficult to monitor the quality of each and every tile manually.
Lot of human resources is required for the defect detection of the tiles. Also it is quite tedious and time
consuming method. Considering this criteria, an automated defect detection and classification technique has
been proposed for the better quality of tiles in manufacturing process as well as production rate. The proposed
method consist three basic steps. The first step is preprocessing the image. Preprocessing operation consists
image acquisition, image enhancement, noise reduction, edge detection. In second step, we applied proposed
flaw detection technique on tiles image to verify whether the tiles is faulty or not.
Keywords: Quality control, Image acquisition, image enhancement, noise reduction, edge
detection,
I.INTRODUCTION
Six types of defects from the defect detection method. These types of defects are shown in the following Table
As the price of ceramic tiles depend on its limpidness and precision of surface texture, color and shape, it’s in
fact a great challenge to control surface eminence and uphold production rate in the field of industrial fabrication
of ceramic tiles. Under consideration these criteria, an enhanced automatic surface flaw detection and
categorization procedure that is able to guarantee the quality of ceramic tiles as well as production rate in
industrial fabrication. The proposed model plays an important role for automatic revealing of surface flaw during
production and packaging. This proposed model includes an automatic categorization technique using computer
vision that helps us to make sense about the pattern of surface defect within a very short time and also helps to
make quick decision about the recovery process. Moreover, it also ensures the quality of tiles automatically
during packaging procedure so that the defected tiles may not be mixed up with the fresh tiles. Objective of this
project is to propose an efficient surface flaw inspection and categorization procedure which will be able to
uncover the surface defects of ceramic tiles at a high rate within a dumpy time. We generally have found total:
127 | P a g e
Table 1. Defects present in ceramic tiles.
Name of defect Description
Crack Breakdown of tile
Blob Water drop spot on tile surface
Pinhole Scattered isolated black-white
pinpoint spot
Spot Discontinuity of color on surface
Edge Break down of edge
Corner Break down of tile corner
Although automated defect detection method have been deployed in ceramic tiles industries since few years but
there still have complex procedure to classify defects using human vision i.e. automated classification and
grading mechanism of packing have not been implemented yet. Again human judgment may be inclined by
anticipation and aforementioned to awareness. In fact, most inexperienced observers have the same opinion that
the flaw may have still there, when they cannot classify the structure of tiles properly. Such a monitoring task is
naturally wearisome, prejudiced and costly in terms of production environment. Objective of this research paper
is to propose an efficient surface flaw inspection and categorization procedure which will be able to uncover the
surface defects of ceramic tiles at a high rate within a dumpy time.
II.BACKGROUND
Since last decade, some defect inspection mechanisms have been proposed to identify the surface flaw of
industrial products (i.e. ceramic tiles, steel bar and wooden surface) by capturing their real time surface image.
Their proposal can be described briefly as follows:
A. Elbehiery et.al. , proposed a method to identify the surface flaw of ceramic tiles. Their proposed method is
divided into two distinct portions. First portion of this method consist with the captured image of tiles as input
and output of this portion is histogram equalized image with intensity adjustment. After that, they use the output
of first portion as input for the second portion. Furthermore, second portion also comprises with different
complementary image processing operations so as to identify and to classify a variety of surface and structural
defects. Their proposed system is not automated rather it emphasizes on the human visual inspection of defect
classification in industrial environment. Moreover, this system is suffered by redundant operation since they
apply the second portion on every test image to identify and classify various types of defects. Thus this system is
time consuming as well.
B. Boukouvalas et al. , they applied separable line filters for flat tiles to identify crack and pinhole defects.
Again, they applied winger distribution for crack detector and a novel conjoint spatial-spatial frequency
representation for textured tiles. In terms of color textured tiles, this type of detection algorithm which looks for
abnormalities both in chromatic and structural properties. However, use of separate filtering technique for
identifying distinct defect is not a good practice.
Consequently, high computational time is taken while we are to handle a large number of operations during
production time. It also proceeds with visual defect classification with human intervention.
128 | P a g e
Se Ho Choi et al. applied a real time mechanism for surface flaw detection of steel coil and bar in high speed
production environment. They used a scheme named “edge preservation” for noise cutback and performance
improvement. In addition, they used “second derivative Laplacian” filter to differentiate gray scale images from
each other. Finally, they applied “double thresholding” technique to formulate binary images. Still, this type of
technique is unable to find the orientation of the edge of surface, because they use “second derivative laplacian”
filter which malfunctions for corner and curves flaw detection as well . In contrast, they hadn’t developed any
automatic classification mechanism rather it was also a human vision process to classify the surface flaw. After
thoroughly revision of previous research paper, there may exists eight types of defects which may occur during
production time and/ or packaging time.
III. METHODOLOGY FOR DEFECT DETECTION AND CLASSIFICATION
3.1 Introduction
A new surface defect detection and classification method has been proposed by this section. Our proposed
method consists of two major portions. One includes some pre-processing image operations to contrast features.
And another portion includes some prominent feature extraction operations to identify defects and to classify
those defects as well. Our proposed model also introduces several algorithms by which we can boost up the
system performance at a higher rate than existing one during production time. Thus it also can reduce the
computational time all together. Here, we applied our mechanisms step-by-step on ceramic tiles image which is
captured before by a digital camera.
3.1.1 First Step: Performing Some Image Pre-Processing Operations
Earlier than applying our proposed defect detection method, initially, we must make use of several image pre-
processing operations on the input images. Pre-processing operations include RGB to Binary conversion, image
enhancement, noise reduction etc. Pre-processing operations are imperative for renovating the captured RGB
image.
A) Image Acquisition
.B) Image Enhancement
Where, O(x,y) represents the output image, I(x,y) represents the pixel position in input image. In this equation,
ni correspond to the number of intensity levels, i stand for the initial intensity level, "min" and "max" represent
the minimum intensity value and the maximum intensity value in the current image respectively. Here "no. of
intensity levels" refers the total number of intensity values that can be assigned to a pixel. For example, normally
in the gray-level images, the lowest possible intensity is 0, and the highest intensity value is 255. Thus "no. of
intensity levels" is equal to 256. The contrast stretching operation is applied on the grayscale images in two
passes. In the first pass the algorithm calculates the minimum and the maximum intensity values in the image,
and in the second pass through the image, the above formula is applied on the pixels. In the proposed method,
129 | P a g e
we enhance the gray level image to improve its visual quality and machine recognition accuracy using the
following formula,
G=INTRANS(F’,stretch’,M,e)
Here, performs the intensity or gray level transformations and G computes a contrast stretching transformation
using the following MATLAB expression:
contrast=1./(1+(M./F+eps)).^E)
Where, parameter M must be in range [0,1]. The default value for M is and the default value for E is 4. Here, F
is gray-level image and M is such result which is found by applying image double and median filtering operation
on F. eps returns the distance from 1.0 to the next largest double-precision number, i.e.
eps=2^(-52)
C) Noise Reduction
a. Consider each pixel in the image.
b. Sort neighboring pixels into order based upon their intensities.
c. Replace the original value of the pixel with the median value from the list.
A median filter is a rank-selection (RS) filter, a particularly harsh member of the family of rank-conditioned
rankselection (RCRS) filters ; a much milder member of that family, for example one that selects the closest of
the neighboring values when a pixel's value is extremely in its neighborhood, and leaves it unchanged otherwise,
is sometimes preferred, especially in photographic applications. Median filter technique is good at removing salt
and pepper noise from an image, and also causes relatively little blurring of edges, and hence is often used in
computer vision applications.
D) Edge Detection
Every edge extraction techniques are consists of two distinct phases:
a. Finding pixels in the image where edges are likely to occur by looking for discontinuities in gradients.
b. Linking these edge points in some way to produce descriptions of edges in terms of lines and curves.
For the proposed method, we detect edge using sobel edge detection method upon the resulting image. Actually
there are many kinds of edge detectors. We use first derivative edge detector (sobel) to detect edges of the
image.
IV. APPLYING THE PROPOSED DEFECT DETECTION PROCESS
All preprocessing operations are applied to the reference image, stored in the computer database to compare with
the test image. Let, the resulting image is I2. Now we consider I1as the resulting image found from the test
image after applying all preprocessing operations. We propose here a new technique. We store I1 and I2 as
matrix form to a file. Let, these two matrices are named as m1 and m2. Then we count the total number of black
pixels (in binary, it is represented as 1) in m1 and that in m2. These two are then compared. If the number of
black pixels in m1 is greater than the number of black pixels in m2 then we can make decision that defect is
found in the test image, otherwise we can say that no defect is present to the test image. To understand this
concept clearly, consider the following example:
130 | P a g e
Let, we have a test image and a reference image of equal size. Now applying preprocessing steps on the test
image we find matrix m1 whose value is:
1 0 0 1 0 1 0 0 0 0
0 0 1 0 0 0 0 1 0 0
m1 = 0 0 0 0 0 m2= 0 0 0 0 0
0 0 1 0 1 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0
Here, number of black pixels for reference image is2 and for the test image is 6. So here ,no. of black pixels
for test image is greater than reference image. So defect is present on the test image.
V. EXPERIMENTAL RESULTS AND DISCUSSION
Defect Detection
The proposed system detects surface flaw both for plain and textured tiles successfully. In this section, represents
the experimental result of our proposed defect inspection technique. We also classify here different types of
defects found through defect detection process. Here it is needed to mention that during the production, many
numbers of tiles are produced in industries at the same time of same colors, shape and pattern. So, all the tiles of
one shape are compared with that particular type of standard tile while processing through the computers. To get
practical realization of our proposed surface flaw detection process, we have applied the proposed procedures on
a sample RGB image of flat tiles. After that we check whether there is any kind of defect exists in this test image
or not by applying our proposed preprocessing operations on this sample RGB image (i.e. image enhancement,
noise reduction and edge detection). This is shown in the following Figure. In this Figure, we also show the
reference RGB image for that test image and the output after applying preprocessing operations on it.
A. Test images
B. Preprocessing Steps on Test Image
Fig 1: Result for reference image Fig 2: Result for test (crack defected) image
131 | P a g e
Fig 3:Result for blob defected image Fig4: Result for blob defected image
Fig 5: Result for spot defected image Fig 6: Result for pinhole defected image
Defects Identification of Computation
defects time in second
Blob Found 1.2429
Crack Not Found 0.7634
Edge Found 0.1693
Corner Not Found 0.1650
Spot Not Found 0.2694
Pinhole Found 0.2642
Table 2: Classification Results
VI. CONCLUSION
The six kinds (crack, blob, pinhole, spot, corner, edge) of defect can be classified using given method .The
proposed method fails to detect the glaze and scratch faults. However, it may be future work to detect and
classify the glaze and scratch faults. We haven’t measure yet the computational time of the proposed
categorization technique. In this case, future work may be calculation of the computational time and provide an
efficient method of reducing computational time for defect classification.
132 | P a g e
REFERENCES
[1] R. C. Gonzalez and R. E. Woods, “Processing”, Pearson Education (Singapore), Pte. Indian Branch, 482
F.I.E. Patparganj, 2005
[2] Puyin Liu and Hongxing Li, “ Theory and Application”. World Scientific
[3] H. Elbehiery, A. Hefnawy, and M. Elewa, “ Defects Detection for Ceramic Tiles Using Image Processing
and Morphological Techniques Proceedings of World Academy of Science, Engineering and Technology,
vol 5, pp 158 1307-6884.
[4] C. Boukouvalas, J. Kittler, R. Marik, M. Mirmehdiand, and M. Petrou, “Ceramic Tile Inspection for
Structural Defects”, under BRITE BE5638, pp 6, University of Surrey, 2006.
[5] Se Ho Choi, Jong Pil Yun, Boyeul Seo, Young Su Park and Sang Woo Kim, “Real Algorithm for High-
Speed Steel Bar in Coil Proceedings of World Academy of Science, Engineering and Technology, Volume
21, January 2007, ISSN 1307 6884.
[6] Mohamed Roushdi, “Comparative Study of Edge Detection Algorithms Applying on the Grayscale Noisy
Image Using Morphological Filter”,Volume 6, Issue 4, December, 2006
[7] Z. Hocenski, T. Keser and A. Baumgartner and efficient method for ceramic tile surface defect
classification.
[8]] Digital Image ”, Ltd., 2005-2006. Fuzzy Neural Network. Scientific, 2004. Surface Techniques”, 158-160,
April 2005, ISSN Colour and ”, BRITE-EURAM, project no. Real-Time Defects Detection Coil”, dings
1307-GVIP Journal, 2006. Baumgartner, “A simple Efficiency (%) detection”, IEEE International
Symposium on Industrial Electronics (ISIE2007), Pp. 1606-1611, 4-7 June 2007.
[9] G. S. Desoli, S. Fiornvanti, R. Fioravanti and D. Corso, “A System for Automated Visual Inspection of
Ceramic Tiles”, Dept. of Biophysical and Electronic Engineering, Via Opera Pia 11/A 16145 Genova,
University of Genoa, Italy.
133 | P a g e
