The doctoral dissertations of the former Helsinki University of Technology (TKK) and Aalto University Schools of Technology (CHEM, ELEC, ENG, SCI) published in electronic format are available in the electronic publications archive of Aalto University - Aaltodoc.
|
|
|
Dissertation for the degree of Doctor of Science in Technology to be presented with due permission of the Department of Computer Science and Engineering for public examination and debate in Auditorium T2 at Helsinki University of Technology (Espoo, Finland) on the 20th of October, 2006, at 12 o'clock noon.
Overview in PDF format (ISBN 951-22-8353-0) [1770 KB]
Dissertation is also available in print (ISBN 951-22-8352-2)
There are two properties which all industrial manufacturing processes try to optimize: speed and quality. Speed can also be called throughput and tells how much products can be created in a specified time. The higher speeds you have the better. Quality means the perceived goodness of the finished product. Broken or defective products simply don't sell, so they must be eliminated.
These are contradicting goals. The larger the manufacturing volumes, the less time there is to inspect a single product, or the more inspectors are required. A good example is paper manufacturing. A single paper machine can produce a sheet of paper several meters wide and several hundred kilometers long in just a few hours. It is impossible to inspect these kinds of volumes by hand.
In this thesis the indexing and retrieval of defect images taken by an automated inspection machine is examined. Some of the images taken contain serious defects such as holes, while others are less grave. The goal is to try to develop automated methods to find the serious fault images from large databases using only the information in the images. This means that there are no annotations. This is called content-based image retrieval, or CBIR.
This problem is examined in two different ways. First the PicSOM CBIR tool's suitability for this task is evaluated. PicSOM is a platform for content-based image retrieval developed at the Laboratory of Computer and Information Science, Helsinki University of Technology. PicSOM has earlier been succesfully applied to various different CBIR tasks.
The other part involves developing new algorithms for efficient indexing of large, high-dimensional databases. The Evolving Tree (ETree), a novel hierarchical, tree-shaped, self-organizing neural network is presented and analyzed. It is noticeably faster than classical methods, while still obtaining good results.
The suitability and performance of both CBIR and ETree on this problem is evaluated using several different experiments. The results show that both approaches are applicable for this real world quality inspection problem with good results.
This thesis consists of an overview and of the following 7 publications:
Keywords: surface inspection, content-based image retrieval, tree-structured self-organizing neural networks, self-organizing maps, Evolving Tree
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
© 2006 Helsinki University of Technology