Incremental learning of a large number of noisy texture classes by similarity measure

Pattern recognition methods are well applied to static problems. Their extension towards incremental acquisition or evolution of class descriptions over time is extremely difficult or even impossible. Moreover pattern recognition methods are not suitable to the integration of numeric and symbolic features in the acquisition of class descriptions. More flexible methods of the acquisition of class descriptions are methods of machine learning technology. For example, learning-from-examples methodology incorporates induction process in order to derive concept descriptions from a given set of preclassified examples. An alternative method that integrates some elements of pattern recognition and machine learning techniques is presented. The proposed method consists of two phases. In the first phase the systems learn a discriminatory description, called, Principle Axes Base (PAB) of initial classes. In the second phase PAB is used to learn new classes