Abstract—Nowadays, the increasing population increases the
consumption as well. Manufacturing systems are developing in
a fast pace to meet increasing demand of consumption. However,
very quick increase of production has surpassed the
development speed of currently existing control systems. In
manufacturing, since the quality is a very important issue as
well as the quantity, the operation of quality control systems
must be accelerated and must be accomplished by machines.
The idea of our study is based on this thinking. Especially defect
detection on widely used glass material that is extremely
difficult to accomplish by humans can be implemented in a
quick, accurate and stable way. In the presented method,
various defects like scratches, bubbles, cracks, corrosion on the
glass surface can be identified. Glass images obtained from a
homogenously illuminated medium are processed by wavelet
transform and obtained images from the wavelet transform are
denoised. Finally, Shannon threshold method is applied to the
images. From the obtained results, defects like scratches,
bubbles on the surface of the glass can be detected successfully.
Index Terms—Glass defect detection, texture analysis, image
processing, wavelet transform, machine vision.
I. INTRODUCTION
Nowadays, as a result of the rapidly expanding automation
systems, it is possible to increase number of manufactured
products in the proportion of supply and demand. In the
globalized world, very fast and effective productions in low
cost are aimed to be prepared for possible demands.
Depending on the arbitrary or compulsory requirements,
glass and glass products are preferred in many different
sectors such as packaging, mirror, kitchen industry,
insulation, heating. Among the manufactured glass products,
defected ones must be prevented to reach to the customer. To
ensure that, the manufactured products must be quality
controlled before leaving the factory. Quality control after
production is usually made by a machine network included to
the manufacturing automation to reduce defects.
Machine-based quality control systems are quite
advantageous compared to the human-based ones [1]-[3].
However, during the search for possible defects on the glass
by sensory technologies, the transparency and light
transmittance of the glass lead to extreme difficulties and
long delays to obtain desired results [4], [5]. For this, the
workload of camera which is image acquisition sensor should
be facilitated. One of the most important points is lighting
system. The lighting system must be designed taking into
account features of the camera. Lighting system should be
effectively illuminated the glass surface. Its main task is to
help image processing unit through illuminating
homogeneously whole glass surface. Other important issue is
position of the camera. It must be correctly calculate size of
defects on the glass surface. Thus, one or more cameras can
be used. If the angle increases between the glass surface and
the camera, calculations and estimating shape of defects
becomes more difficult.
Possible problems encountered on the glass during
production are defects like scratches, bubbles, cracks and
fractures. When these problems are detected during quality
control procedure, the products can be recycled and
reprocessed before they leave the factory. However, if the
defected glass reaches the consumer, besides increasing the
cost of recycling, the factory may be discredited. Therefore, it
is clear that the quality control must be made very carefully
and the product must be presented to the customer after this.
In this study, by implementing the quality control of glass in
an effective way, an effort is made to increase the quality of
the manufactured glass and avoid possible financial losses. A
number of different methods have been employed in glass
defect analysis and many studies for effective control are
present in literature. In order to detect the surface defects on
the glass, Markov random field [6] Otsu threshold method [7],
canny edge detection method [8], binary feature histogram
method [9], Fuzzy C Means clustering algorithm [10] have
been employed. In this study, the images taken from
homogenously illuminated medium are first preprocessed to
be denoised, processed via wavelet transform and
transformed into binary form by Shannon method. Shannon
Method is a common method to evaluate the edges [11]-[13].
Therefore, the defects over the glass can be detected in a
quick and accurate manner.
Wavelet analysis is becoming a popular tool for analyzing
localized variations of power within a time series. By
decomposing a time series into time–frequency space, one is
able to determine both the dominant modes of variability and
how those modes vary in time and wavelet transform has
been used for numerous studies such as image processing,
geophysics, ocean waves etc. [14]-[16].
In this study, the taken images have been evaluated using
wavelet transform for surface detection. Wavelet improves
Shannon entropy evaluates. Wavelet and Shannon entropy
were used together to determine the edges.
Fig. 1. Image processing unit.
Rest of this paper is organized as follow; Section II is
mentioned about system overview and definition of problem.
Section III is explained wavelet transform and using this
Glass Surface Defects Detection with Wavelet Transforms
Bayram Akdemir and Şaban Öztürk
170
International Journal of Materials, Mechanics and Manufacturing, Vol. 3, No. 3, August 2015
Manuscript received December 4, 2014; revised March 31, 2015. This
project (114E925) is supported by TUBİTAK.
The authors are with the Electrical and Electronics Engineering
Department, Engineering Faculty, University of Selcuk, Konya, Turkey
(e-mail: {bayakdemir, sabanozturk}@selcuk.edu.tr).
DOI: 10.7763/IJMMM.2015.V3.189