$ 11 A.... LANDSAT 
: m... NIMBUS 
34 
8 A 
iz ae À 
33 1.] a MAN y E ro Suan 
ia 0 
  
0 5 lo 15 
Number of control points per 2.106 pixels 
Figure 3: Rectification accuracy of satellite scan imagery, obtained 
by various authors under different conditions. 
Procedures for this task have been extensively described in the literature 
(Koneeny, 1976). The most common are indirect or "resampling" methods, 
where a new, rectified image is produced,starting from the regular raster 
of pixels of the rectified image. The gray value for each pixel is found 
by looking for the corresponding location in the unrectified image. The gray 
value encountered in that location or a function of the surrounding gray 
values is attached to the pixel in the rectified image. 
3.3 Image Enhancement 
We understand "enhancement" to be an operation to make non-obvious in- 
formation that is presented in an image more obvious to someone visually 
interpreting it. 
(a) Contrast enhancement is by far the most commonly applied image enhance- 
ment procedure. It consists of an intensity transformation which maps 
brightness values in the input image to other values in the output image. 
Several techniques are in common use. The simplest is a linear contrast 
enhancement in which some image density L is mapped to black, some value 
H is mapped to white, and values in between are scaled proportionately. 
Values extreme of L and H are saturated to black and white respectively. 
L and H can either be found manually be inspection of the histogram, or 
automatically by a program which saturates pre-specified percentages of 
the image histogram to black and white. 
Non-linear intensity transformations can also be used. In their most 
general form, these consist of a table assigning an output intensity value 
for each possible input value. Again the transformation can be manually 
specified based on inspection of the histogram, or a particular transfer 
function, e.g. logarithmic or power law, could be applied. 
Alternatively, after scanning the input histogram, a program can specify 
a tabular transformation which will produce an output histogram approximating 
a predetermined shape, such as a uniform or Gaussian distribution. Forcing 
a uniform distribution results in the greatest contrast enhancement being 
applied to the most populated range of brightness in the input image. This 
property mekes the uniform distribution stretch particularly useful as a 
quick look evaluation procedure. Principal difficulties are that it is 
sometimes too harsh, and tends to compress and hence loose information 
in the light and dark "tails" of the histogram. Definition of detail at 
the extremes of the histogram tends to be preserved or increased more 
     
  
  
  
  
  
  
   
   
   
   
   
    
  
  
  
  
  
   
   
  
  
   
    
   
  
  
  
   
  
  
  
F 
tn 
(D h^ ck Fr or 
OFF HOO H