Feature extraction by multilayer perceptron: visualization of internal representation of input data
Miloš Oravec, Jarmila Pavlovičová · 2007
In this paper, we present methodology for manipulating the internal representation of input images formed by Multilayer Perceptron (MLP). Methods of ordering hidden layer outputs to two-dimensional form are described. Next, reordering of such representation is shown. Presented methods can be applied in feature extraction for image compression and pattern recognition.