In SMAP, spectral signatures are extracted from images based on training map by determining the parameters of a spectral class Gaussian mixture distribution model, which are used for subsequent segmentation (i.e. classification) of the MS images. The Gaussian mixture class describes the behavior of an information class which contains pixels with a variety of distinct spectral characteristics. For example, forest, grasslands or urban areas are examples of information classes that need to be separated in an image. However, each of these information classes may contain subclasses each with its own distinctive spectral characteristic; a forest may contain a variety of different tree species each with its own spectral behavior. Mixture classes improve segmentation performance by modelling each information class as a probabilistic mixture with a variety of subclasses. In order to identify the subclasses, clustering is first performed to estimate both the number of distinct subclasses in each class, and the spectral mean and covariance for each subclass. The number of subclasses is estimated using Rissanen's minimum description length (MDL) criteria [28]. This criteria determines the number of subclasses which best describe the data. The approximate maximum likelihood estimates of the mean and covariance of the subclasses are computed using the expectation maximization (EM) algorithm [29-30].

SMAP improves segmentation accuracy by segmenting the image into regions rather than segmenting each pixel separately [31-32]. The algorithm exploits the fact that nearby pixels in an image are likely to have the same class and segments the image at various scales or resolutions using the coarse scale segmentations to guide the finer scale segmentations. In addition to reducing the number of misclassifications, the algorithm generally produces segmentations with larger connected regions of a fixed class. The amount of smoothing that is performed in segmentation is dependent on the behavior of the data. If the data suggest that the nearby pixels often change class, then the algorithm adaptively reduces the amount of smoothing, ensuring that excessively large regions are not formed (http://wgbis.ces.iisc.ernet.in/grass/grass70/manuals/html70_user/i.smap.html).