The program NEM clusters a given data set, using the Neighborhood EM algorithm proposed by Ambroise 1996. All parameters are to be specified on the command line.
NEM computes a partition of a given set of objects described by one or several numeric variables. Some of the variables may be the geographic position of the object. This algorithm is derived from the EM algorithm applied to mixture distributions. Its new feature consists in taking into account spatial interdependance between the objects.
The algorithm takes as input an objects-variables table, and a specification of the neighborhood relationship between the objects. It produces as output a fuzzy or a hard partition of the objects.
NEM may be used for:
- unsupervised segmentation of color or gray-level images (points = pixel values, geographic position = pixel coordinates) ;
- clustering of spatial data like socio-economical activities of neighbouring counties, etc.