A comparative study of ART2-A and the self-organizing feature map

This paper compares the ART2-A model and the self-organizing feature map. The two models are applied to the classification of feature vectors extracted from texture images. Simulation shows that ART2-A performs best when its noise-reduction/contrast-enhancement mechanism is switched off. In this mode it performs better than the self-organizing feature map. Experiments known from literature show that a backpropagation network performs only slightly better than ART2-A for the same texture classification task.