Publications Details
A template-based approach to clustering
A new approach for solving two-dimensional clustering problems is presented. The method is based on an inhibitory template which is applied to each pair of dots in a data set. Direct clustering of the pair is inhibited (allowed) if another dot is present (absent), respectively, within the area of the template. The performance of the method is thus entirely determined by the shape of the template. Psychophysical experiments have been used to define the template shape for this work, so that the resulting method requires no pattern-dependent adjustment of any parameters. The novel concept of a psychophysically-defined template and the absence of adjustable parameters set this approach apart from previous work. The useful grouping performance of this approach is demonstrated with the successful grouping of a variety of dot patterns selected from the clustering literature.