Pattern Recognition using Kohonen Featuremap

This is an example of a kohonen network

by Udo Lugauer, Stefan Grosse & (obviously not) Walter Berger

It will recognize the capital Letters

A B C D E F or H


At first you have to chose the learnsteps ('number of cycles')
of the network (100, 500, 1000 or 2000).

Pressing the 'learn' button starts the training phase.
The three diagrams show you the learnrate the error and
the featuremap during the learn-phase

(The 'reset button' re-initialises the network.
you have to train the network again, if you press this button!!)

After the Network has learned the captal Letters, you can
try writing any of these letters (capital A,B,C,D,E,F,H) in the white paintbox
using your mouse.

Here’s an example of some letters that should be recognized:

Now let's see if the network recognizes the correct letter... (press 'recognize')
hopfully it was a success...
pressing 'clear' sweeps the paintbox - now you can try again.

back to main page