A new algorithm for the implementation of clustering

In pattern recognition applications sometimes we need to classify pattern samples under the circumstances of lacking priori knowledge, such as the number of classes, the distribution of classes, etc. or having a complicated distribution of classes in the characteristic space. A new clustering algorithm is presented to realize clustering with an unknown number of classes and with an unknown or complicated distribution of classes. When the number of classes is known, this algorithm is also effective and has a good adaptability for various distributions of classes