Přístupnostní navigace
E-application
Search Search Close
Publication detail
ŠKORPIL, V. ŠŤASTNÝ, J.
Original Title
Kohenen Neural Network for Image Processing. In Proceeding of the 10th International Workshop on Systems, Signals and Image Processing
Type
conference paper
Language
English
Original Abstract
The Kohonen network is one of the self-organizing neural networks. The number of inputs coming to neurons is equal to the number of inputs to the Kohonen network. A characteristic feature of an ART network is its ability to switch the variable and stable mode without damaging the information learned. The ART network is very sensitive to noise and failure and that is why it is not suitable for the recognition of damaged pictures. It is not necessary to solve the problem of local minimum as the ART network is always convergent to optimum solution. A real technological scene was simulated by digitizing two-dimensional pictures of real objects. The ART network gave relatively good results. Its application is suitable especially for the recognition of not very damaged and noisy pictures and in the case when other patterns are added to the learned network. The Kohonen network presented excellent results. The learning went off very quickly and that is why the Kohonen network is suitable for applications where the patterns are often changed. When tested, the Back-propagation neural network the same as the Kohenen network yielded excellent results. With this network, the learning took longer than with the Kohenen network, but the results of identification of the chosen damaged patterns were better.
Keywords
neural network; Kohenen Network; image processing
Authors
ŠKORPIL, V.; ŠŤASTNÝ, J.
Released
26. 8. 2003
Publisher
CVUT Praha
Location
Praha
Pages from
183
Pages to
188
Pages count
5
URL
knihovna UTKO
BibTex
@inproceedings{BUT32455, author="Vladislav {Škorpil} and Jiří {Šťastný}", title="Kohenen Neural Network for Image Processing. In Proceeding of the 10th International Workshop on Systems, Signals and Image Processing", booktitle="Proceeding of the 10th International Workshop on Systems, Signals and Image Processing", year="2003", pages="183--188", publisher="CVUT Praha", address="Praha", url="knihovna UTKO" }