Structural Pattern Recognitionby Non - Reducible

The classiication technique by Non-Reducible Descriptors provides decision rules in term of Boolean formulae on object's key features. In this paper we present an extension of the Non-Reducible Descriptors that contains quantitative information. By Quantitative Non-Reducible Descriptor of an object, the distance to the descriptions of objects from another classes is explicitly expressed. This fact can be used in the supervised pattern recognition problems to determine the weight of a descriptor. In the case of the unsupervised pattern recognition problem, we propose a hierarchical procedure based on the concept of Quantitative Non-Reducible Descriptor that maximizes the discrimination between the clusters of the training objects. The Quantitative Non-Reducible Descriptors construction procedure suggested here has two major properties: it has a diminished number of computational operations with respect to the procedure for Non-Reducible Descriptors construction and allows incremental learning of the set structure.