Image classification by extended certainty factors

摘要

Two supervised classifiers based on Certainty Factors (CFs) are described; in particular, a new algorithm for the generation of the base of classification rules is proposed. Such an algorithm specifies what “events” should be used as conditions and with what CFs rules may assign samples to classes. The main novelty lies in the definition of events as one-dimensional adaptively computed functions. Interesting features of the proposed classifiers are the use of comprehensible classification criteria and the automatic “knowledge acquisition” from training patterns. Experimental results obtained in the case of multisensorial remote-sensing images are reported.