Fig. 9: Jos area classification results: 
(a) by the Maximum Likelihood method, and 
(b) by the Separating Hyperplanes method 
  
Class Variance Level 
  
1, reject pixels (pure 
white patches) - 
  
2. Lake (not pure white) 18,9 
3. farmland (light grey) 3.4 
4. bare ground (dark grey) 21.6 
5, built-up areas (black) 5.2 
  
  
  
  
Table 3: Variance levels of objects in the Jos area 
  
  
  
(a) (hb) (a)-(b) 
Farmland 60.05 70.72 -10.67 
Bareground 31.98 13.42 18.56 
  
  
  
  
  
Table 4: Farmland and bare ground as pecentages of the 
whole classification area: 
(a) Maximum Likelihood classificatjon 
(b) Separating Hyperplanes classification 
ur 5. CONCLUSION 
It has been shown that the Maximum Likelihood method of digital image clas- 
sification cannot be used as a "black box" for the classification of all kinds 
Ris of MSS data. The outcome of any classification by this method depends, not only 
$54 Fi on the quantity and quality of the training samples, but also on the inter- 
one action of the differently structured data of the different object classes. An 
single attempt to improve upon the Maximum Likelihood algorithm, through an adjustment 
as ho- of the variance levels of the classes, and reported in a previous paper 
(Ekenobi 1982), yielded only about 50 % higher accuracy. 
69 
LEM i ns OO a v AREE em