Pattern Recognition, 4th Edition

12.3. Sequential Clustering Algorithms

In this section we describe a basic sequential algorithmic scheme, (BSAS), (which is a generalization of that discussed in [Hall 67]), and we also give some variants of it. First, we consider the case where all the vectors are presented to the algorithm only once. The number of clusters is not known a priori in this case. In fact, new clusters are created as the algorithm evolves.

Let d(x,C) denote the distance (or dissimilarity) between a feature vector x and a cluster C. This may be defined by taking into account either all vectors of C or a representative vector of it (see Chapter 11). The user-defined parameters required by the algorithmic scheme are the threshold of dissimilarity Θ and the maximum ...