C. Wemmert and P. Gançarski (France)
Machine Learning, Multi-strategical unsupervised classifi cation, Image processing
Within the framework of our research on unsupervised classification, we were interested in the definition of a generic model of unsupervised classifiers and data to clas sify. Our goal is to be able to handle in the same system several entities of classifiers in order to make them collab orate, without having to know their type. Moreover, we wish to be able to handle data of different types (numeri cal, symbolic or structured). We present here how the classifiers collaborate and how we defined a structure in the data types allowing to handle the objects to classify without knowing their real type. Then we show haw this approach were integrated in a new system of unsupervised multi-step multi-strategical classification, the MUSTICOS system. Finally we show an application of this unsupervised multi-strategical classification system on the problem of thematical urban zones extraction from remote sensing im ages.
Important Links:
Go Back
