Application of Morphological Operators to Supervised Multidimensional Data Classification

Abstract

In this paper the use of mathematical morphology (MM) operators in the several steps of supervised multidimensional data classification methods is discussed. In a first step, that can be called low-level feature extraction, morphological operators are usually applied with filtering purposes but can also be applied to extract topological features that can be used to generate extra information enhancing the phenomena representation. In a second step, MM operators can also be used to to improve pixel based classification algorithms in the following phases: i) modelling of clusters hulls training samples, ii) defmition of criteria for processing overlapping clusters hulls and iii) 2D pattern space partition and distance function definition. Finally, in a third and last step of the classification procedure, low-level generalization algorithms can be used to perform the spatial arrangement of primitive image objects, giving a smoother, more simplified view of the classified image. Two examples of the application of this methodology to supervised classification are presented. The first one concerns the interpretation of geochemical data in a region in the south of Portugal, while the second one concerns the identification and mapping of the areas occupied by different forest cover types.