Improved colour image vector quantisation by means of self-organising neural networks

The problem of colour quantisation is important in many respects: colour monitors can usually display only a number of contemporary colours, and some images need to be represented in an approximate, even if satisfying, way. This is particularly true for the dissemination of images through communications networks and to information terminals. Moreover, colour quantised images can be stored in less space. The colour quantisation algorithm introduced by the authors is based on a set of neural cells structured in a self-organising two-dimensional map. The proposed technique provides high quality images and its neural architecture makes the algorithm flexible even if different kinds of source image are used. The hardware implementation is easy to realise and gives real-time performance