Fig. 8: Hannover area classification results: 
(a) by the Maximum Likelihood method, and 
(b) by the Separating Hyperplanes method 
where only one or two good training pixels are available. The Maximum 
Likelihood requires a minimum of n+1 training pixels per class. 
(n is the number of spectral bands). 
(2) Several object types may be grouped together into a single class. This ST 
would not work with the Maximum Likelihood method in which, infact, the 9f th 
histograms of the classification data should be examined before classifi- actio 
cation to make sure that each spectral band of each class has only one attem 
"peak". The grouping together of more than one object type into a single of th 
class may not be intentional, since an object class is usually not as ho- (Eken 
mogeneous as one would want it to be. 
68 
La \ = ri 3