Novel Technique To Combine Bws And Fws Classifiers

A novel approach to combining the binary weighted and frequency weighted schemes (BWS and FWS) for pattern classification is described. The method presented employs a well established statistical procedure and offers good performance. The principal advantages of the technique are that only a single layer is required and it avoids the serious problem of choosing optimal values for the zero-frequency components which occur in the FWS. Preliminary experiments demonstrate that the choice of operational parameters required is straightforward. Furthermore, the technique is intrinsically parallel and lends itself naturally to parallel implementation.