Pattern classification using neural network based image processing

Pao Yue-kong Library Electronic Theses Database

Pattern classification using neural network based image processing

 

Author: Wan, To-ki
Title: Pattern classification using neural network based image processing
Degree: M.Sc.
Year: 1995
Subject: Optical pattern recognition
Neural networks (Computer science)
Image processing -- Digital techniques
Pattern recognition systems
Computer vision
Hong Kong Polytechnic University -- Dissertations
Department: Multi-disciplinary Studies
Dept. of Manufacturing Engineering
Pages: xii, 180 leves : ill. ; 30 cm
Language: English
InnoPac Record: http://library.polyu.edu.hk/record=b1479356
URI: http://theses.lib.polyu.edu.hk/handle/200/1777
Abstract: Biologically inspired computing is different from conventional computing. A neural network is an information processing system that is non-algorithmic, non-digital, and intensely parallel[Maureen,1992]. With the development of artificial neural network technology, the network has been well considered as a good candidate for pattern classification. The adaptive behaviour of the network makes it very attractive to manufacturing industries such as automating robots and control systems, controlling production line processes, inspecting for quality control, and selecting parts on an assembly line. However, the complexity of the network topology compounded with the massive parallel processing nature, hardware realisation of the network is quite expensive, at least, for commercial use. This study demonstrates the practical approach in software simulation of the network and the implementation in the PC based Microsoft Windows environment. Counterpropagation network, which is a three layers hybrid neural network model and was introducted by Robert Hecht-Nielsen in 1986[James,1991], is implemented in this project using "C" language and Microsoft Visual Basic language as a basic network unit called subnet. The initial weights set up for the neurons are crucial for the network performance and the rate of network convergence. Instead of assigning initial weight's value of the neurons randomly, this report describes the strategic training method for the network to speed up the learning process under limited training samples. One-Class-One-Network approach is used to build the system. Each class of objects is associated with a network that composes with different subnets for different types of feature. To maintain the stability of the network in the weight updating, training is only localised for the each subnet. A decision logic is developed to control and select the final winner. The project focuses on implementing of the neural network technology in low-cost image pattern classification system. The system is trained for twelve quantized orientations of objects in each class. The classification of the object is determined by scoring the activation of the subnets for the three different types of feature vectors, namely, radii signature vectors, boundary(shape) descriptive vectors and moment descriptive vectors. A set of network analysis tools are developed to facilitate the evaluation of the network performance. The performance measurement is based on the network activities and the relative response of the neurons inside the network. A case study is included to illustrate the functionalities and capabilities of the system.

Files in this item

Files Size Format
b14793568.pdf 6.329Mb PDF
Copyright Undertaking
As a bona fide Library user, I declare that:
  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

     

Quick Search

Browse

More Information