SchapeNet

This applet performs pattern recognition on drawn figure, which can be rectangle, triangle or circle. It uses fourier transform in preprocessing phase and neural network later to classify patterns.

Neural network is initialy loaded with pretrained weights, those weights can be loaded any time by pressing "Default NN" button.
Most of the parameters are self explanatory,

"mi" - is momentum coefficient and

"psi" is lerning coefficient.

"Descriptors" parameter
specifies how many values from DFT will be taken. Value 5 indicates that 10 complex values will be taken, which makes input to neural network of size 20. Each descriptor is defined as value from DFT vector DFT_VEC. First descriptor is defined as two values : DFT_VEC[0] and
DFT_VEC[M-1]. Where M is DFT vector length. First value from Fourier transform is zeroed to gain placement independenc.

**Applet is around 60kb also pressing "prev test", "next test" or "train" requires downloading test or training images what can take some
time, please be patient :)**

Marcin Jędrzejewski - 2004