Fast Image Analysis using Kohonen Maps

Authors:

D.Willett, C.Busch, F.Seibert

Keywords:

Artificial Neural Networks, Image Classification, Kohonen Feature Map, Nearest Neighbor Search

Abstract:

The paper considers image analysis with Kohonen Feature Maps. These types of neural networks have proven their usefulness for pattern recognition in the field of signal processing in various applications. The paper reviews a classification approach, used in medical applications, in order to segment anatomical objects such as brain tumors from magnetic resonance imaging (MRI) data. The same approach can be used for environmental purposes, to derive land-use classifications from satellite image data. These applications require tremendous processing time when pixel-oriented approaches are chosen. Therefore the paper describes implementation aspects which result in a stunning speed-up for classification purposes. Most of them are based on geometric relations in the feature-space. The proposed modifications were tested on the mentioned applications. Impressive speed-up times could be reached independent of specific hardware.

Published:

Proceedings of the IEEE Workshop NNSP, pp. 461-470, (1994)

A PostScript version, compressed with gzip, can be found here.


Last modified: Wed June 25 1997 11:39:34

For comments or questions about this page, please send me email.

Christoph Busch