Ebook: Supervised and Unsupervised Pattern Recognition: Feature Extraction and Computational

Издательство CRC Press, 2000, -367 pp.This volume describes the application of supervised and unsupervised pattern recognition schemes to the classification of various types of waveforms and images. An optimization routine, ALOPEX, is used to train the network while decreasing the likelihood of local solutions. The chapters included in this volume bring together recent research of more than ten authors in the field of neural networks and pattern recognition. All of these contributions were carried out in the Neuroelectric and Neurocomputing Laboratories in the Department of Biomedical Engineering at Rutgers University. The chapters span a large variety of problems in signal and image processing, using mainly neural networks for classification and template matching. The inputs to the neural networks are features extracted from a signal or an image by sophisticated and proven state-of-the-art techniques from the fields of digital signal processing, computer vision, and image processing. In all examples and problems examined, the biological equivalents are used as prototypes and/or simulations of those systems were performed while systems that mimic the biological functions are built.
Experimental and theoretical contributions are treated equally, and interchanges between the two are examined. Technological advances depend on a deep understanding of their biological counterparts, which is why in our laboratories, experiments on both animals and humans are performed continuously in order to test our hypotheses in developing products that have technological applications.
The reasoning of most neural networks in their decision making cannot easily be extracted upon the completion of training. However, due to the linearity of the network nodes, the cluster prototypes of an unsupervised system can be reconstructed to illustrate the reasoning of the system. In these applications, this analysis hints at the usefulness of previously unused portions of the spectrum.
The book is divided into four parts. The first part contains chapters that introduce the subjects of neural networks, classifiers, and feature extraction methods. Neural networks are of the supervised type of learning. The second part deals with unsupervised neural networks and fuzzy neural networks and their applications to handwritten character recognition, as well as recognition of normal and abnormal visual evoked potentials. The third part deals with advanced neural network architectures, such as modular designs and their applications to medicine and threedimensional neural networks architectures simulating brain functions. Finally, the fourth part discusses general applications and simulations in various fields. Most importantly, the establishment of a brain-to-computer link is discussed in some detail, and the findings from these human experiments are analyzed in a new light.
All chapters have either been published in their final form or in a preliminary form in conference proceedings and presentations. All co-authors to these papers were mostly students of the editor. Extensive editing has been done so that repetitions of algorithms, unless modified, are avoided. Instead, where commonality exists, parts have been placed into a new chapter (Chapter 4), and references to this chapter are made throughout.
As is obvious from the number of names on the chapters, many students have contributed to this compendium. I thank them from this position as well. Others contributed in different ways. Mrs. Marge Melton helped with her expert typing of parts of this book and with proofreading the manuscript. Mr. Steven Orbine helped in more than one way, whenever expert help was needed. Dr. G. Kontaxakis, Dr. P. Munoz, and Mr. Wei Lin helped with the manuscripts of Chapters 1 and
3. Finally, to all the current students of my laboratories, for their patience while this work was compiled, many thanks. I will be more visible—and demanding—now. Dr. D. Irwin was instrumental in involving me in this book series, and I thank him from this position as well. Ms. Nora Konopka I thank for her patience in waiting and for reminding me of the deadlines, a job that was continued by Ms. Felicia Shapiro and Ms. Mimi Williams. I thank them as well.Section I — Overviews of Neural Networks, Classifiers, and Feature Extraction Methods—Supervised Neural Networks
Classifiers: An Overview
Artificial Neural Networks: Definitions, Methods, Applications
A System for Handwritten Digit Recognition
Other Types of Feature Extraction Methods
Section II Unsupervised Neural Networks
Fuzzy Neural Networks
Application to Handwritten Digits
An Unsupervised Neural Network System for Visual Evoked Potentials
Section III Advanced Neural Network Architectures/Modular Neural Networks
Classification of Mammograms Using a Modular Neural Network
Visual Ophthalmologist: An Automated System for Classification of Retinal Damage
A Three-Dimensional Neural Network Architecture
Section IV General Applications
A Feature Extraction Algorithm Using Connectivity Strengths and Moment Invariants
Multilayer Perceptrons with ALOPEX: 2D-Template Matching and VLSI Implementation
Implementing Neural Networks in Silicon
Speaker Identification through Wavelet Multiresolution Decomposition and ALOPEX
Face Recognition in Alzheimer’s Disease: A Simulation
Self-Learning Layered Neural Network
Biological and Machine Vision