Pattern recognition and the nondeterminable affine parameter problem

Abstract

This thesis reports on the process of implementing pattern recognition systems using classification models such as artificial neural networks (ANNs) and algorithms whose theoretical foundations come from statistics. The issues involved in implementing several classification models and pre-processing operators - that are applied to patterns before classification takes place - are discussed and a methodology that is commonly used in developing pattern recognition systems is described. In addition, a number of pattern recognition systems for two image recognition problems that occur in the field of image matching have been developed. These image recognition problems and the issues involved in solving them are described in detail. Numerous experiments were carried out to test the accuracy and speed of the systems developed to solve these problems. These experiments and their results are also discussed.