Automated Inspection of Defects in Glass by proper Color space ...

Automated Inspection of Defects in Glass by proper Color space selection and 

Segmentation Technique of Digital Image processing 

Nishu #1 

#1 ME ECE Student, UIET 

Panjab University, Chandigarh, India 

nishu.31187@gmail.com 

Abstract 

Glass defects which result into poor quality are a major 

reason of embarrassment for manufacturers. It is an 

extremely tedious process to manually inspect very large 

size glasses. The manual inspection process is slow, timeconsuming 

and prone to human error. Automatic inspection 

systems using image processing can overcome many of 

these disadvantages and offer manufacturers an 

opportunity to significantly improve quality and reduce 

costs. 

In this paper we propose an evaluation system for detection 

of various glass defects like scratches, inclusions, surface 

defects etc. The process involves the color space selection 

and a segmentation technique often used in medical 

imaging has been tested for detection of various defects in 

glass. The color space conversion helps in the selection of 

best color space for quantifying the visibility of the defects 

while the segmentation algorithm involves the use of 

contour method based on intensity inhomogeneities. 

Keywords: Color spaces, defect detection, glass, image 

segmentation. 

1. Introduction 

The quality control concept is the most vital aspect of 

the glass manufacturing industry. In the past human vision 

has played a primary role in quality inspection and 

verification processes. It is, however, now considered a 

limiting factor in the inspection of products coming out 

from modern industrial production lines, where high 

working speeds and very limited tolerances are required, 

unlike traditional defect detection mode which is slow and 

prone to errors. The solution to these problems has been the 

introduction of artificial vision-based inspection system. As 

a matter of fact, applications of these systems are nowadays 

widespread in many industrial sectors [1], particularly the 

glass industry. For this industrial sector, an in-line 

automated inspection system that is able to discover, and 

IJCTA | MAY-JUNE 2012 

Available online@www.ijcta.com 

Nishu et al ,Int.J.Computer Technology & Applications,Vol 3 (3), 1058-1063 

Sunil Agrawal #2 

#2 Assistant Professor, UIET 

Panjab University, Chandigarh, India 

ISSN:2229-6093 

classify the defects present in glass sheets has been 

developed and analyzed [2]. 

The quality control of the final product is a 

fundamental part of the glass production process, and this is 

demonstrated by the considerable scientific research that 

has been devoted to automatic inspection techniques [3]. 

These studies focus on using different approaches for 

defect detection depending on their specific application 

because no single technique can be considered optimal. As 

a result, many inspection techniques have been proposed 

with the aim of increasing the productivity and improving 

the final product quality [4]. 

With regard to the glass industry, analyses and 

methodologies employed to detect the defects in the glass 

sheets mainly use image processing techniques because of 

their higher precision and speed [5]. A number of 

techniques that use machine vision defect detection system 

have been presented in the past, by various authors in their 

research on this topic. 

The rest of the paper is organized as follows: Section 

two discusses about the various types of glass defects. 

Section three briefly reviews the work related to detection 

of defects. Section four introduces color space conversion 

test used for quantifying the visibility of various defects in 

different color spaces and the segmentation method using 

contours applied for detection of defective areas in glass 

sheets. Section five discusses the results and concludes the 

paper. 

2. Types of Defects 

Once the glass sheet is manufactured, it is sent to the 

defect detection division of the glass production unit for 

testing and validation of defects. The various types of 

defects that can be present in the glass are: 

a) Foreign material: This defect has the appearance of a 

lump. It is an unmelted, opaque material embedded in 

the glass. For example: Dust, sand particles, chemical 

residues. 

1058

) Discoloration Defects: These defect areas are roughly 

defined as fairly large, several millimeters in diameter, 

and relatively dark and/or bright regions that stand out 

against the background [6]. Example: Water marks, 

which occur during heating and annealing, blackening 

or red color of glass which is caused due to heating. 

c) Line Defects: These are the marks or irregular patches 

on the surface [7]. These occur mainly during 

transportation within the factory. These can be light 

(like the marks made by using some tools) or deep 

(penetrating into the surface of an item and can be felt 

on touching the surface). These can occur during the 

process of edge grinding and corner cutting. Example: 

Scratches and spots, knot line. 

d) Edge defects: Edge defects are the main cause of glass 

breakage during its production. They can be prevented 

by detecting them at an early stage and rejecting 

suspicious sheets. Production line uptime will increase; 

production costs will be reduced accordingly. 

Example: jagged edges. 

e) Point Defects: These are the inclusions trapped inside 

glass as a defect during its production. Example: 

Bubbles, stone, melt inclusions. 

f) Surface defects: These are the surface defects which 

cause major problems for manufacturers, particularly 

when the production process includes a surface 

treatment stage. Example: Holes and dirt. 

Different image processing algorithms are required for 

the detection of different types of defects which have been 

reviewed in the next section. 

3. Related Work 

There has been considerable research in the field of 

defect detection in glass utilizing in-line automated 

inspection system. Makoto, Akira and Toshio [8], in their 

paper, proposed a method for detecting foreign materials in 

the inspection of an LCD with its protective film in place. 

The surface of the LCD is scanned under a fan-beam laser 

light to obtain a set of light-section time-series images. 

These images are composed into a horizontal cross-section 

image of the specified depth and internal foreign materials 

are detected from it. To detect the low-contrast regions on 

glass, a highly robust estimator, known as the Model- 

Fitting (MF) estimator [9] was developed by X. Zhuang et 

al. which used a modified MF estimator to robustly 

estimate the background model and as a by-product to 

segment the blemish defects, the outliers. The illumination 

irregularity was made as a parabolic function; the center 

area was made brighter than the perimeter of the image. A 

zero mean Gaussian noise was added to the ground truth 

and the amount of noise, the standard deviation of the 

Gaussian noise, and the depth of circle of the ground truth 

are controlled for each simulation. A system [10] was 

designed to reproduce the real issues of an in-line quality 




ISSN:2229-6093 

control system which included three subsystems: an array 

of several CMOS cameras, a controllable roller conveyor, 

and a PC-based image processing system that is also 

responsible for the control of the other subsystems. The 

detection of the defects was performed by means of canny 

edge detection, with thresholds chosen according to some 

statistics of the images being processed. Jie Zhao, Xu Zhao 

and Yuncai Liu [11] proposed a method for detection of 

bubbles and inclusions. First, the defect region was located 

by the method of canny edge detection, and thus the 

smallest connected region (rectangle) was found. Then, the 

binary information of the core region was obtained based 

on an OSTU [12] and an adaptive algorithm. After noises 

were removed, a Binary Feature Histogram (BFH) was 

used to describe the characteristic of the glass defect. 

Finally, the AdaBoost method was adopted for 

classification. 

The quality requirements for glass have continuously 

increased over the past years. We propose an algorithm 

which would be able to detect a defective area in glass with 

reasonable accuracy. The design and implementation of the 

algorithm is carried out using MATLAB. This would offer 

manufacturers with an opportunity to significantly improve 

quality and reduce costs. Significant and clear images of 

the different glass defects would enable production staff to 

trace back to the cause of the defect without delay ensuring 

high product quality. 

4. Defect Detection Process 

Defects are complicated and uncertain. According to 

appearing areas, defects can be separated into various 

categories. There are three main modules of this process. 

Those are image preprocessing module, color space 

selection module and image segmentation module. The 

image preprocessing module basically involves formatting 

of the images in the database as per the requirement to 

achieve the best results at later stages of research. Next step 

is to select a color space which best shows the defect in an 

image thereby reducing the further complexities during its 

detection during segmentation stage. 

4.1. Color space conversion 

The selection of color space is one of the determinants 

of the image segmentation quality; the segmentation results 

would be more accurate if an appropriate color space is 

adopted. During this process each input image containing a 

defect is converted from RGB to four other color spaces 

given below: 

4.1.1 RGB 

In RGB color space, the colors red, green, and blue are 

mapped onto a 3-D Cartesian coordinate system which 

results is a 3-D cube. The vertices of cube are the primary 

colors (red, green, and blue) and the secondary colors 

(cyan, yellow, and magenta) of light. 

1059

4.1.2 HSV 

The HSV color model is a kind of method to define 

colors according to the three basic features of the color: 

Hue, Saturation and luminance [13].Hue is the color type 

which ranges from 0 to 360.It is expressed as an around a 

color hexagon, using the red axis as the reference (0˚) axis. 

Saturation, the vibrancy of color ranges from 0 to 100% 

and is also called as the „purity‟. It is measured as the 

distance from the V axis. Value is basically the brightness 

of the color that ranges from 0 to 100%. It is measured 

along the axis of the cone. 

4.1.3 YCbCr 

It is a model in which Y is the intensity component. Cb 

refers to blue color component and Cr refers to the red 

color component [14]. This color model is basically used in 

digital videos. 

4.1.4 NTSC 

In this type of color format, image data consist of three 

components: luminance (Y), hue (I) and saturation (Q), 

where the choice of letters YIQ is conventional [15]. The 

luminance component carries the g ray-scale information 

and the other two components carry the color information. 

4.1.5 Gray Scale 

In photography and computing, a grayscale digital 

image is an image in which the value of each pixel is a 

single sample, that is, it carries only intensity information. 

Images of this sort, also known as black-and-white, are 

composed exclusively of shades of gray, varying from 

black at the weakest intensity to white at the strongest. 

Once the conversion is done, each color space is 

assigned a level from 1 to 5, known as Five point Likert 

scale, according to the degree of visibility of defects in the 

input image in each color space that objectively agrees with 

human visual perception. The color space with highest 

degree of visibility of the defect is marked 5, with a lower 

degree as 4 as so on and the one with least degree of 

visibility is assigned as 1. Thus, evaluation corresponds, on 

average, with the assessment of a group of inspectors. The 

conversion time of each image from RGB color space to 

other color spaces is also calculated. 

Figure 1 Color Space Conversion Analyses 




4.2. Image Segmentation 

Image segmentation is the key step of the process from 

image processing to image analysis. The quality of 

segmentation effect influences the follow-up analysis of the 

images directly [14]. Therefore accurate segmentation of 

image is very essential. The purpose of image segmentation 

is to divide the image into a number of significant regions 

based on some characteristics (intensity inhomogeneities 

here [16]), making these characteristics to display similarity 

in single region and display difference between different 

regions. 

In the process of detecting defects in glass, region based 

active contour model in a variational level set formulation 

has been implemented for segmentation (proposed by 

Chunming Li, Chiu-Yen Kao, John C. Gore, and Zhaohua 

Ding [16] ) which uses intensity inhomogeneity as a region 

descriptor to identify the region of interest that is to be 

segmented. First of all a non-negative kernel function K is 

defined which is chosen to be a gaussian kernel given as: 

with a scale parameter σ > 0. 

If C be a closed contour of the image Ω, Ω1 is the region 

outside(C) and Ω2 be the region inside (C) then for point 

x ε Ω local intensity fitting energy is given as: 

where are positive constants 

ISSN:2229-6093 

are the values that approximate 

intensities in Ω2 and Ω1 regions of image 

I(y) are the intensities in the local region centered around 

point x 

K(x-y) is the weight assigned to the each intensity at y 

(1) 

(2) 

is the weighted mean square error of the image 

intensities I(y) by fitting values . 

This region scalable fitting energy function whose size is 

controlled by the Gaussian kernel is defined in terms of 

contour and approximates the intensities of the image on 

either side of the contour. Now to obtain the complete 

defect boundary a contour C is found that minimizes the 

above found energy for all x in image Ω which is done as 

below: 

(3) 

Where│C│is length of the contour penalized to smoothen 

the contour. 

1060

As defined in [16], to minimize the energy functional given 

by eq. 4, its gradient flow is used as level set evolution 

equation 

(4) 

where φ is the level set function and takes positive and 

negative values outside and inside the contour C 

respectively 

H is the Heaviside function 

The gradient flow equation is given as: 

(10) 

Equation 11 minimizes the energy functional for a fixed 

value of φ. 

(5) 

(6) 

(7) 

(8) 

(9) 

(11) 

5. Results 

The proposed method has been applied to digital 

images of the glass sheets taken for detecting various 

defects like scratches, inclusions, surface defects etc using 

MATLAB. The test has been conducted on 83 images 

containing various kinds of defects that can be present in 

the glass sheets. Some typical results and discussion are 

given below. The input image is converted to four color 

spaces specified above. Two factors taken into account for 

analysis are: 

Five-Point Likert scaling 

Conversion Time 

According to the Likert scale, level from 1 to 5 is 

assigned to each color space and the analysis below shows 

that RGB color space gives the best results with level 5 

being assigned for all the images of the database. The 

second best color space according to human perception of 




degree of defect visibility is Gray color space and then 

Ntsc, HSV and last comes the YCbCr color space. 

From time analysis graph shown below minimum time 

for conversion is taken by RGB-HSV color space while the 

order of time taken for conversion from minimum to 

maximum is Gray, Ntsc and YCbCr color space. 

So combining both the results, it can be concluded that 

the defects are best visible in RGB color space and RGB – 

Gray color space conversion would give the best results 

when these defects are detected in glass industries. 

Figure 2 Five Point Likert Scale Analyses 

Figure 3 Conversion Time Analyses 

ISSN:2229-6093 

This RGB image obtained is then given as input image to 

the segmentation algorithm implemented. The input image 

is read and converted to double for better precision. The 

various parameters taken into account for use in different 

equations were initialized. The original image was scaled 

and converted to gray color map. The convolution of the 

image with the Gaussian kernel and the convolution of the 

Gaussian kernel with constant 1 were computed. The above 

mentioned calculations were then done step by step for the 

level set evolution. The defect boundary was determined 

collectively for every 20 iterations. The final contour is 

marked at the end of the 100 iterations. The image is 

finally inverted and gives the clear defect boundary on the 

basis of the intensity inhomogeneity on both sides of the 

contour. The complete results have been show in figure 4. 

1061

Figure 4Surface Defect Detection 

Figure 5Scratch Detection 

Figure 6Inclusion Defect Detection 




ISSN:2229-6093 

6. Conclusion 

In this paper various defects present in glass sheets have 

been briefly discussed. Various color spaces have been 

analyzed and compared for their performance in 

quantifying the visibility of defects in digital images. On 

the above reported theory and results the RGB color space 

is recommended if one needs to view and process various 

defects like scratches, spots, edge defects etc in images. If 

any color space conversions are required the best choice is 

RGB to Gray color space conversion which makes the 

defect visible to largest extent. The segmentation algorithm 

using the contour method which has been often used in 

medical imaging has been test in the glass industry to detect 

the various defects. 

Our future work will focus on: 1) Testing the technique on 

bulk of defective images as this test was done on a limited 

database. 2) Classification of defects according to their 

characteristic features. 3) Using more computational 

resources to improve the efficiency of defect detection 

techniques. 

References 

[1] E.N. Malamas, E.G.M. Petrakis, M. Zervakis, L. Petit, 

J.D. Legat, “A survey on industrial vision systems, 

applications and tools, Image and Vision Computing”, 

2003, pp. 171–188. 

[2] F. Adamo, F. Attivissimo, A. Di Nisio, M. Savino, “An 

Automated visual inspection system for the glass industry”, 

In: Proc. Of 16th IMEKO TC4 Symposium, Florence, Italy, 

Sept 2008, pp. 22–24. 

[3] M.A.Coulthard, “Image Processing for Automatic Surface 

Defect Detection,” Surface Inspection Ltd, UK, pp.192-196. 

[4] M. Leconte, “Laser glass inspection system”, International 

Society for Optical Engineering, 1997, pp.878–882. 

[5] Zhang Yepeng, Tao Yuezhen, Fan Zhiyong, “Application of 

Digital Image Process Technology to the Mouth of Beer 

Bottle Defect Inspection,” 2007, pp. 2-905- 2-908. 

[6] Nishu Gupta, Sunil Agrawal, “Glass Defect Detection 

Techniques using Digital Image Processing –A Review”, 

IJCA Special Issue on IP Multimedia Communications, 

October 2011, pp. 65-67. 

[7] www.wikipedia.org 

[8] Makoto Shimizu, Akira Ishii and Toshio Nishimura, 

“Detection of Foreign Material Included in LCD Panels,” 

IEEE conf 2000, pp.836-841. 

[9] Chang-Hwan Oh, Hyonam Joo and Keun-Ho Rew, 

“Detecting Low-Contrast Defect Regions on Glasses Using 

Highly Robust Model-Fitting Estimator,” International 

Conference on Control, Automation and Systems, 2007, pp. 

2138-2141. 

[10] Francesco Adamo and Mario Savino, “A low-cost Inspection 

system for online defects assessment in satin glass,” 2009, 

pp. 1304-1311. 

[11] Jie Zhao, Xu Zhao and Yuncai Liu, “A Method for Detection 

and Classification of Glass Defects in Low Resolution 

Images,” Sixth International Conference on Image and 

Graphics, 2011, pp.642-647. 

1062

[12] Peng X, Chen Y and Yu W, “An online defects inspection 

method for float glass fabrication based on machine vision,” 

International Journal of Advanced Manufacturing 

Technology, vol.39, 2008, pp.1180-1189. 

[13] Li Shuhua, Guo Gaizhi, “The application of improved HSV 

color space model in image processing”, 2nd International 

Conference on Future Computer and Communication, 2010. 

[14] Wang Yuedong, Xue Heru, “Studies on Color Space 

Selection and Methods of Segmentation Quality Evaluation”, 




ISSN:2229-6093 

1st International Conference on Information Science and 

Engineering, 2009. 

[15] Digital Image Processing Using MATLAB by Rafael C. 

Gonzalez and Richard Woods, second edition. 

[16] Chunming Li, Chiu-Yen Kao, John C. Gore, and Zhaohua 

Ding, “Minimization of Region-Scalable Fitting Energy for 

Image Segmentation”, IEEE Transaction on Image 

Processing, Vol. 17, No. 10, October 2008, pp. 1940-1949. 

1063

Automated Inspection of Defects in Glass by proper Color space ...

Create successful ePaper yourself

Delete template?

Save as template?