Digital Realisation of Self-Organising Maps
Nigel M Allinson, Martin J. Johnson, K.J. Moon · Neural Information Processing Systems · 1988
A digital realisation of two-dimensional self-organising feature maps is presented. The method is based on subspace classification using an n-tuple technique. Weight vector approximation and orthogonal projections to produce a winner-takes-all network are also discussed. Over one million effective binary weights can be applied in 25ms using a conventional microcomputer. Details of a number of image recognition tasks, including character recognition and object centring, are described.