Figure 1: Generation of a digital image 
re, the existing moisture and other factors. The surface 
reflectance properties are normally expressed in the so 
called bidirectional reflectance distribution function 
(BRDF). 
For reasons of simplicity one distant point light source 
illuminating the object surface with constant radiance 
from the direction 5 is introduced only. Also, the object 
surface is assumed to look equally bright from every 
viewing direction. This assumption is equivalent to Lam- 
bertian reflection, except that light absorption at the 
object surface is allowed here. The ratio between inco- 
ming and reflected radiant flux - a value between 0 and 
1 - is called the albedo and is denoted by p ( X, Y. In 
this case L can be written as 
ns (2) 
|n | 
  
1 
Leisp(Xy) 
E,(X,Y) scene irradiance (constant) 
p (X,Y) albedo of the object surface 
5 unit vector in the direction of illumi- 
nation at P ( X , Y, Z ) 
n vector in the direction of the object 
surface normal at P( X, Y, Z ) 
Combining equations (1) and (2) yields: 
83 0) 
In| 
Er) stud EC 
  
834 
In the sensor an image intensity value g(x,y) - in 
general an integer value between 0 and 255 - is recorded 
rather than the image irradiance E; (x,y). g (x,y) is 
proportional to E; (x , y ) : 
8(x,y) 9 k E:(x,y) (4) 
g(x,y) image intensity value at P' (x , y ) 
k rescaling constant 
All constants of equations (3) and (4) can be combined 
with the albedo into the so called object intensity value 
G(X,Y): 
(5) 
4 
cosa |. d.» 
7 (FV TEp(X.Y) 
G (X,Y) object intensity value atP (X,Y, Z) 
G(X,Y) =k 
  
Substituting equations (3) and (5) into (4) yields: 
ns (6) 
|n | 
  
g(x,y) = G(X.,Y) 
3. DIGITAL IMAGE MATCHING 
The algorithms for digital image matching are usually 
classified into three groups: 
- image matching using signal processing algorithms 
(also called area based image matching), 
- feature based image matching, 
- relational image matching. 
In the first group a function of the intensity value diffe- 
rences between selected windows of the different images 
is minimized. The maximization of the well known cross 
correlation coefficient /Hannah 1989/ as well as the least 
squares matching algorithms /Fórstner 1982; Grün 1985; 
Rosenholm 1986/ and phase shift methods /Ehlers 1983/ 
all belong to this first group. The algorithms of the 
second group search for predefined features (points, 
edges, lines, regions) independently in the images /Bar- 
nard, Thompson 1980/. Low level image processing al- 
gorithms are employed for the selection of the features. 
Based on the output of these algorithms a list of possibly 
corresponding features is established. This list still con- 
tains a number of gross errors and ambiguities and is 
thinned out using for instance robust estimation or dy- 
namic programming. The famous zero-crossing algo-