Geography Reference
In-Depth Information
Fig. 5.36 Combine the 14-class illustrated in the previous figure in one thematic map, using
ArcGIS-software to sum the individual thematic results, and LCCS-software to prepare the
legend
MLC
The Maximum Likelihood Classifier (MLC) has been employed since the late
1940 s. It found increasing investment in the two fields of: pattern recognition; and
remote sensing techniques (Nilsson 1965 ). It is offered in about all remote sensing
and image processing software packages, and it is usually applied as the typical
supervised classification approach. It is a widely robust supervised algorithm, and
it is the primary approach for most multi-spectral remote sensing interpretations at
present (Lillesand et al. 2008 ). Its general concept defines the maximum likelihood
decision rule, which is the probability that a pixel belongs to an individual class.
This classifier is derived from the Bayes-rule in which classes have equivalent
priorities. It uses the training data gathered during field-work or on image itself to
calculate the mean vector and variance-covariance matrix for each required class.
Both means and variances are then employed to assess the probabilities (Jensen
2005 ). This algorithm is based on the supposition that the likelihood degree
function for each class is multivariate, and often a Gaussian distribution is
assumed. A pixel is lastly classified to that class, for which it has the highest
probability (Lillesand et al. 2008 ).
MLC operates (see Fig. 5.37 ) by using the training-samples-based means and
standard deviations of individual spectral bands in order to scheme LULC classes
as centroids in feature space. These centroids are circumscribed by likelihood
Search WWH ::




Custom Search