LVQ and Kohonen Nets as Human, for Comparing ASM Generated Faces

Porncharoensin, Hataikan (2002) LVQ and Kohonen Nets as Human, for Comparing ASM Generated Faces. MSc(R) thesis, University of Glasgow.

Full text available as:
[thumbnail of 13818746.pdf] PDF
Download (3MB)

Abstract

This thesis has three main parts; Artificial Neural Networks, the Active Shape Models, and an illlustration of combining them. Computer Calculations are in the Mathematica Language. The first part is between Chapter I and 4. Neural Networks can be used in many areas as shown in some examples in Chapter 1. At the beginning, some simple nets are used to solve logic functions (AND, OR). More details about AND, OR problems are in Chapter 2. This chapter is concentrated on pattern classification; Hebb Net and Perceptron. In Example 2.8, the net gave the output of Character E as both E and K. Therefore, this problem is solved by using a competitive net in Chapter 3 to select the winner. Many examples are shown in this chapter because different topological structures of a Kohonen net give different results. In addition, a Kohonen net can be applied to solve a Traveling Salesman Problem. Chapter 4 gives the details of Backpropagation Neural Nets. The output propagates the error back to the previous hidden layer, it then calculates the error, and updates the weights. The backpropagation net can be used to compress data as well. In Chapter 5, the general idea of Active Shape Models is given. This method can be used to generate new shapes, which are similar to the original shapes yet of good variety. The selected shapes are hands and faces. Finally, this method is used in Chapter 6 where we compare Learning Vector Quantization, Kohonen nets and human observation, for grouping hand and face shapes.

Item Type: Thesis (MSc(R))
Qualification Level: Masters
Additional Information: Adviser: Stuart Hoggar
Keywords: Applied mathematics, Artificial intelligence
Date of Award: 2002
Depositing User: Enlighten Team
Unique ID: glathesis:2002-75970
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 19 Nov 2019 17:10
Last Modified: 19 Nov 2019 17:10
URI: https://theses.gla.ac.uk/id/eprint/75970

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year