Preserving visual perception by learning natural clustering

The neural clustering behavior of self-organizing neural networks enables the learning of perceptually meaningful pattern features and makes it possible to store pictorial data in an effective way. The authors experiments show that the storage of perceptual features requires a fraction of the size of the original data, and still renders little or no difference compared with the original. Experimental results of natural clustering and non-trivial clustering from corner-propagation networks using feature map and frequency-sensitive variations of the Kohonen network are shown and discussed