Skip to main content
Fachgebiete
Automobil + Motoren Bauwesen + Immobilien Business IT + Informatik Elektrotechnik + Elektronik Energie + Nachhaltigkeit Finance + Banking Management + Führung Marketing + Vertrieb Maschinenbau + Werkstoffe Versicherung + Risiko
DE
EN
Bücher
Zeitschriften
Themenseiten
Organisationspsychologie Projektmanagement Marketing Smart Manufacturing
Weitere Formate
Podcasts Webinare Technik Webinare Wirtschaft Kongresse Veranstaltungskalender Awards
Jetzt Einzelzugang starten
Zugang für Unternehmen
Referenzkunden
SLX-Digitalkonferenz: Zukunftswerkstatt 2024

Mehr Umsatz mit Top-Kontakten

Am 7. Mai erfahren Sie, wie Vertriebsteams Leadgenerierung, Leadnurturing und Leadstrategien effizient steuern können.
Erweiterte Suche
Anmelden
nach oben

1998 | Buch

Introduction and Theory Kapitel lesen Erstes Kapitel lesen

Image Processing using Pulse-Coupled Neural Networks

verfasst von: Thomas Lindblad, PhD, Jason M. Kinser, DSc

Verlag: Springer London

Buchreihe : Perspectives in Neural Computing

Enthalten in: Professional Book Archive

Einloggen, um Zugang zu erhalten
insite
SUCHEN

Über dieses Buch

Pulse-coupled neural networks represent a new and exciting advance in image processing research. When exposed to grey scale or colour images they produce a series of binary pulse images which allow the content of the image to be assessed much more accurately than from the original. In this volume Thomas Lindblad and Jason Kinser provide a much needed introduction to the topic of PCNNs. They review the theoretical foundations, and then look at a number of image processing applications including segmentation, edge extraction, texture extraction, object identification, object isolation, motion processing, foveation, noise suppression and image fusion. They also look at the PCNNs ability to process logical arguments and at how to implement it in specialised hardware. It will be of particular interest to researchers and practitioners working in image processing, especially those involved with medical, military or industrial applications. It will also be of interest to graduate-level students.

Inhaltsverzeichnis

Frontmatter
Chapter 1. Introduction and Theory
Abstract
Humans have an outstanding ability to recognise, classify and discriminate objects with extreme ease. For example, if a person was in a large classroom and was asked to find the light switch it would not take more than a second or two. Even if the light switch was located in a different place than the human expected or it was shaped differently than the human expected it would not be difficult to find the switch. Another example is that of recognising dogs. A human needs to see only a few examples and then he is able to recognise dogs even from species that he has not seen before. This recognition ability also holds true for animals, to a greater or lesser extent. A spider has no problem recognising a fly. Even a baby spider can do that. At this level we are talking about a few hundred to a thousand processing elements or neurons. Nevertheless the biological systems seem to do their job very well.
Thomas Lindblad, Jason M. Kinser
Chapter 2. PCNN Theory
Abstract
The Pulse-Coupled Neural Network is to a very large extent based on the Eckhorn model except for a few minor modifications required by digitisation. The early experiments demonstrated that the PCNN could process images such that the PCNN output was quite similar for images that were shifted, rotated, scaled, and skewed. Subsequent investigations determined the basis of the working mechanisms of the PCNN and led to its eventual usefulness as an image-processing engine.
Thomas Lindblad, Jason M. Kinser
Chapter 3. PCNN Image Processing
Abstract
Traditional image processing is a vast and extensive field covering many different approaches. We will not attempt to cover the broad extent of this science here. However, it is important to understand some of the fundamentals of image processing so that comparisons to the PCNN can be made. Traditional image processing is manifested in many forms but all based on a limited set of fundamental processing operations (e.g. convolution of matrices).
Thomas Lindblad, Jason M. Kinser
Chapter 4. The PCNN Kernel
Abstract
The PCNN convolution kernel is one of the main components of the PCNN. It can be manipulated to provide a variety of computations. The original Eckhorn model used a Gaussian type of interconnections, but when the PCNN is applied to image processing problems these interconnections are available to the user for altering the behaviour of the network.
Thomas Lindblad, Jason M. Kinser
Chapter 5. Target Recognition
Abstract
Target recognition is the ability to find and delineate objects within an image. This is not an easy task for most applications. The targets can be presented in a variety of different views, rotations, scales, illumination, etc. Traditional target recognition methods have been successful only in limited cases and views. However, the three basic steps to Automatic Target Recognition or ATR are well known: (i) segment the image into regions, (ii) determine features in the image regions, and (iii) classify the regions by their features. Living creatures do this ail the time. Flies do it. Fish do it. Spiders do it, even little spiders. Yet it is still eluding researchers.
Thomas Lindblad, Jason M. Kinser
Chapter 6. Dealing with Noise
Abstract
The pulses produced by the PCNN are consistent when the input scene is shifted, rotated, scaled, etc. In other words, the PCNN can still segment an object under these conditions. The one image attribute that causes problems for the PCNN is noise. In the presence of non-trivial noise, neurons lose the ability to synchronise their pulses.
Thomas Lindblad, Jason M. Kinser
Chapter 7. Feedback
Abstract
Recent findings in the investigations of the rat’s olfactory bulb demonstrated an inhibitory feedback mechanism [Ambros-Ingerson, 90]. It is generally thought that this mechanism categorises the input during iterations so that the input information transitions from coarse data to fine data in a hierarchical manner. In other words, the rat broadly classifies the smell at first and then the inhibitory connections subtract the coarse nature of the signal. The remaining finer signals provide for more discriminatory classification.
Thomas Lindblad, Jason M. Kinser
Chapter 8. Object Isolation
Abstract
Object isolation is the ability to isolate an object (target) within a frame. In other words, the intensity of the target increases while the intensity of non-targets decreases. By the end of the process only the target remains in the field of view.
Thomas Lindblad, Jason M. Kinser
Chapter 9. Foveation
Abstract
The human eye does not stare at an image. It moves to different locations within the image to gather clues as to the content of the image. This moving of the focus of attention is called foveation. A typical foveation pattern [Yarbus, 65] is shown in Figure 9.1. Many of the foveation points are on the corners and edges of the image. More foveation points indicate an area of greater interest.
Thomas Lindblad, Jason M. Kinser
Chapter 10. Image Fusion
Abstract
In a multi-spectral environment information about the presence of a target is manifest across the spectra. Detection of these targets requires the fusion of these different kinds of data. However, image fusion is difficult due to the large volume of data. Typically each detector channel does not provide enough information to detect the target with a significant level of confidence. Thus, each channel provides clues only and hints as to the presence of the target.
Thomas Lindblad, Jason M. Kinser
Chapter 11. Software and Hardware Realisation
Abstract
The PCNN is a powerful algorithm with many applications. This chapter is concerned with the actual implementation of the PCNN in software and hardware. Implementing the PCNN in computer code is not very difficult, but a straightforward implementation may result in a code for which the runtime is not very fast. On a 90 MHz Pentium a single PCNN iteration for a 128x128 image takes about one second. While this may be sufficient in many cases, it is not fast enough for real-time applications. Therefore, implementation in parallel computers, dedicated hardware and optical computers are also discussed.
Thomas Lindblad, Jason M. Kinser
Chapter 12. Summary, Applications and Future Research
Abstract
In the previous chapters we have seen a variety of tasks that the pulsecoupled neural network can perform. These typically involve the tasks of image processing, enhancing features, removing noise, etc. Below we will summarise these capabilities, including some features that were not touched upon in the previous chapters and include some examples.
Thomas Lindblad, Jason M. Kinser
Backmatter
Metadaten
Titel
Image Processing using Pulse-Coupled Neural Networks
verfasst von
Thomas Lindblad, PhD
Jason M. Kinser, DSc
Copyright-Jahr
1998
Verlag
Springer London
Electronic ISBN
978-1-4471-3617-0
Print ISBN
978-3-540-76264-5
DOI
https://doi.org/10.1007/978-1-4471-3617-0