8/9/2019 Localization of Designed Texture on Glass Bottles

1/51

Localization of Designed Texture on Glass Bottles Page 1

Technical report, IDE1032, June 2010

Localization of designed texture

On glass bottles

Master Thesis in Intelligent Systems

Abhishek Sharma Pallav Nandi Chaudhari

850326-1250 850908-T314

School of Information Science, Computer and Electrical Engineering

Halmstad University

8/9/2019 Localization of Designed Texture on Glass Bottles

2/51

Localization of Designed Texture on Glass Bottles

8/9/2019 Localization of Designed Texture on Glass Bottles

3/51

Localization of Designed Texture on Glass Bottles

Preface

This treatise is a masters thesis for the Master of Science program in intelligent systems at

the Halmstad University, Sweden in the IDE department. This thesis deals with the

localization of designed texture on glass bottles at the hot end. This is a research issue from

the company Gedevelop located at Helsingborg, Sweden. The work is carried out with the

guidance of Mr. Kenneth Nilsson. The data for the experiments are provided by Gedevelop.

This project is implemented in the Matlab programming language on a personal computer

system and experimental results on the proposed method are presented.

8/9/2019 Localization of Designed Texture on Glass Bottles

4/51

Localization of Designed Texture on Glass Bottles

8/9/2019 Localization of Designed Texture on Glass Bottles

5/51

Localization of Designed Texture on Glass Bottles

Abstract

Hot end inspection of glass bottles necessitates the localization of designed textures

on that bottle. In this paper we proposed a method to localize designed textures on a test

bottle by matching its image to a certain number of reference images and then mapping the

texture to the test bottle. The intended purpose of this algorithm is to localize the designed

texture on the bottle so that the designed texture areas and other areas on the image of the

bottle can be distinguished. The localization is done by finding the amount of rotation

caused due to the motion of the conveyor belt. The average orientation angle feature is

used in the matching. Reference databases are created which contain necessary information

regarding the different textured regions. A selected portion of the image is used to match

the bottle at the hot end with the same area of the reference bottles. The design texture is

marked, according to the appropriate match. All the methods involved experiments and

results and conclusions based upon them are presented subsequently.

8/9/2019 Localization of Designed Texture on Glass Bottles

6/51

Localization of Designed Texture on Glass Bottles

8/9/2019 Localization of Designed Texture on Glass Bottles

7/51

Localization of Designed Texture on Glass Bottles

Contents

1 Introduction 1

2. Background.... 4

2.1 Real world scenario and their impact.. 5

2.2 Texture recognition and localization..7

2.2.1 Direct template matching7

2.2.2 Feature based matching7

2.3 Propose methods.. 8

2.3.1 Reference model 8

2.3.2 Computations on test image.9

2.3.3 Limitations. .10

2.3.3 Hot spot marking.11

3. Method12

3.1 Feature Extraction12

3.1.1Dominant local direction...13

8/9/2019 Localization of Designed Texture on Glass Bottles

8/51

Localization of Designed Texture on Glass Bottles

3.2 Skew detection and correction and target area

extraction. 15

3.2.1 Skew detection. 15

3.2.2 Target area extraction..16

3.3 Similarity Measure..22

3.4 Hot spot marking.23

4. Experiments and results..24

4.1 Software environment..24

4.2 Data24

4.3Feature Extraction28

4.4 Registration.31

4.5 Matching32

4.6 Hot Spot Marking.36

5. Conclusions..38

References.40

8/9/2019 Localization of Designed Texture on Glass Bottles

9/51

Localization of Designed Texture on Glass Bottles

8/9/2019 Localization of Designed Texture on Glass Bottles

10/51

Localization of Designed Texture on Glass Bottles

1

1 IntroductionFault detection is an essential part of the glass manufacturing process. Different types of

faults occur in glass containers during the manufacturing process [9]. The glass containers

are always inspected by automatic systems. Sometimes human inspection is also done [14].

Most of the inspection is done at the cold end of the manufacturing process. Cold end is a

relatively time consuming process and fault detection is an independent part of this phase.

The company Gedevelop[11] is looking for some method which can detect faults at the hot

end phase, because hot end inspection may give some advantages over regular cold end

inspection as an early warning of problems. Gedevelop is a system development company

which develops and maintains systems for glass manufacturing plants. They have their

camera system which can take pictures of bottles at the hot end and they are interested in

finding manufacturing defects. For detection of faults by the analysis of images, it is

necessary that the system must be able to differentiate between designed texture and

textures made out of defects. One precise way of doing this is by localizing the designed

texture on the glass containers. Once it islocalized, it will be easier to find the areas on the

bottle where inspection should be made for defects.

Hot end phase puts several constraints on the task of fault detection. The environment at

the hot end is not a very friendly environment in which to perform an experiment.

Traditional fault detection equipments cannot work there; no kind of handling of the bottles

is possible. One possible way of inspection is vision based systems. There are also some

limitations imposed by the situation. Multiple numbers of cameras cannot be used. The glass

containers come very fast on the conveyor belt from the molding machine and there is very

8/9/2019 Localization of Designed Texture on Glass Bottles

11/51

Localization of Designed Texture on Glass Bottles

2

limited gap available between two bottles on the conveyor belt. Space is very limited for

fitting multiple numbers of cameras so, normally, only one camera can look at a bottle at a

certain time.

In vision based systems, one of the possible ways to detect different kinds of faults

e.g. cracks, foreign inclusions, etc, can be inspected by texture analysis. In that case, there

must be a classification methodwhich can identify the designed texture on the bottles, so

that the designed texture and textures made out of faults can be differentiated.

In the identification of designed texture on glass bottles at the hot end, there are

certain factors which, to a certain extent, provide the opportunity for it to be modeled in an

easier way, because of its conceptual relevance to some very efficient texture recognition

algorithms, such asoptical character recognition [2, 12]. On the other hand, the real world

situation makes it difficult by imposing some constraints upon the image capturing task.

Practically, when the bottle comes onto the conveyor belt from the forming machine, the

camera can look only at one side of the bottle, which implies that the whole texture area on

the bottle will not be visible to the camera. Due to the high speed of the belt, the bottle may

be rotated to a certain degree. Because of this rotation, the same texture area will not

always be captured by the camera.Keeping in mind all these points, here, we propose a

method to localize the designed texture by finding the rotation angle of the bottle by

matching it to certain number of references and then mapping thetexture to the test bottle.

The method is passive in nature, because it does not directly extract the designedtexture.

In our proposed method, the image pixels are represented by their average directions

extracted from linear symmetric features [4] in a local neighborhood. Rather than using

Matching procedure to find the texture directly, we used matching to find the amount of

rotation. After finding the amount of rotation with respect to a reference image, the

designed texture is marked according to the matching reference.

8/9/2019 Localization of Designed Texture on Glass Bottles

12/51

Localization of Designed Texture on Glass Bottles

3

This treatise is composed as follows:

In chapter 2, the background, an overview of the real world situation in the present

context and motivation for the proposed procedures, is explained. In chapter 3, the

methods involved are discussed in detail, whereas in chapter 4 and chapter 5, the

experiments and their results, and their future scopes, are described.

8/9/2019 Localization of Designed Texture on Glass Bottles

13/51

Localization of Designed Texture on Glass Bottles

4

2 BackgroundTexture recognition is an image processing technique which has a wide variety of

applications in different scientific research areas e .g. optical character recognition[2], face

recognition[15] fingerprint recognition[1][8], biometrical identification of laboratory

animals[13], medical image analysis[6], robot navigation [16]etc. For human eyes it is