Scale-based approach to hierarchical fuzzy clustering. (Q1575341)

Sensorial signals are processed by the brain by relying on their significant aspects. Fuzzy and scale-based approaches try to imitate this mechanism. In the paper, a new clustering algorithm is proposed which makes use of both approaches. It is characterised by a hierarchical splitting process guided by the scale-based approach and based on the repetitive application of an improved version of the Min--Max fuzzy algorithm. In each iteration of the algorithm at least one cluster is split and a scale parameter is determined. The optimal partition is decided based on a stability criterion defined as a function of the scale. Several tests illustrate the performance of the algorithm, also in the framework of video databases management systems. In fact, hierarchical clusters of video frames seem to be very appropriate for browsing a video sequence, especially if they are determined by a scale-based criterion simulating different resolution levels of the human observation. Moreover, fuzzy sets play a fundamental role because of the resulting soft decision criteria in the critical task of scene change detection.