The suitability of layer-based systems for fuzzy information 
representation in GISs has been much debated [Hadzilacos, 
1994]. For the specific problem at hand, they are expected to 
be highly effective. Photogrammetric accuracy indices are 
typically addressing points or lines. From such indices, 
object positioning accuracies can be composed, in the same 
manner that complex objects are generated by combining 
points and linear elements, permitting the substitution of 
deterministic object outlines by more accurate probabilistic 
layers. Fig. 3 displays this concept, with an object outline 
(left) and its probabilistic representation (right). The 
probabilistic representation describes the existence of the 
object at various field locations, with field values ranging 
e.g. between 0 (white, lack of object) and 1 (black, definite 
existence of object). The figure shows how an error in 
measuring the building’s top left corner (e.g. due to 
obstructions by other objects, shadows or other unfavorable 
radiometric conditions) is affecting the accuracy of the 
complete building object. 
  
  
  
  
Fig. 3: An object outline (left) and its probabilistic 
representation (right) 
This recording of accuracy estimates is very essential when 
considering the availability of databases of various scales 
and accuracies within information systems. It allows us to 
move towards ‘objective GIS’, where biases caused by 
unaccounted uncertainties in data are avoided, and thus, the 
subjectivity of data accuracies is removed. 
5. EFFECTS ON DIGITAL PHOTOGRAMMETRY 
The expected benefits of image analysis operations through 
their embedding into integrated — photogeographic 
environments are related to accessing the geoinformation 
which already exists for the scenes they depict. Currently, 
typical image analysis operations aim at the automation of 
certain photogrammetric tasks by addressing imagery 
isolated from the environment in which these operations are 
performed. By performing these operations in an 
information-supported mode (Fig. 4), algorithmic progress 
can be greatly accelerated. 
The logical operations which can be performed using 
information extracted through photogrammetric digital 
image analysis are of dual nature: object- or scene-oriented, 
with the difference between the two referring to the view of 
specific information alone or within a broader context. 
Object-pertinent logical operations are mostly consistency- 
oriented, and can be of geometric (e.g. examining whether 
the corners of an extracted building outline are orthogonal) 
or semantic (e.g. examining whether a complex extracted 
building outline actually represents muliple rather than a 
single building) nature. In this sense, they can be viewed 
rather as extensions of the individual digital image analysis 
28 
28 
modules addressing object globality, as they enhance these 
modules by ensuring local coherency, 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
New 
Imagey 
Geoinformation 
ly Database e 
Spatial & 
Semantic 
Query Operations 
v | 
Automated Suitable Existing 
Operations on Geoinfo for 
New Imagery € Area in New 
Imagery 
a 
| ! Fi enema ! | 
I 
New Geolnfo —— 3 | Logical Operations 
  
  
  
  
  
  
Fig. 4: Conceptual diagram of digital image analysis ope- 
rations and information flow in an integrated environment 
It is the database-supported scene-oriented operations 
though which are expected to have the most profound effect 
on digital photogrammetric operations. They consist of 
logical tests employing broader scene information, and their 
purpose is to examine whether the newly acquired 
geoinformation is consistent with the already existing one 
(e.g. is it possible for a road outline to be substantially 
higher than the local DEM values?). The role of such 
operations should not be considered limited to consistency 
checks, but rather it should be viewed as providing links to 
or triggering additional automated operations (note the 
interupted arrow in Fig. 4). The extraction of an object 
which has been modified since its last record provides leads 
for other potential changes. New road segments for example 
can signal local changes in buildings or land use and vice 
versa. Thus, suitable algorithms should be employed to 
examine whether these tied-in occurancies have indeed taken 
place, and if so to record their effects on geoinformation. In 
a similar manner, apparent logical inconsistencies can 
trigger additional image analysis operations to verify or 
corect the extracted geoinformation. The satisfaction of 
logical conditions permits the updating of the 
geoinformation database. 
The digital photogrammetric research issue which is 
expected to be most improved by integrated 
photogeographic environments is the automated extraction 
of objects from digital imagery, which is a fundamental GIS- 
oriented application. By employing previous database 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
inform 
qualita 
extract 
minim 
sought 
expect 
operati 
expect 
digital 
relevai 
The su 
bringii 
geogra 
advanc 
becom 
an inte 
revolu 
extrac 
traditi 
well a 
contex 
(throu, 
evolve 
image. 
than s 
QU Ac 
Aerial 
Conve 
Q Ag 
for Ai 
of Ge 
L Ag 
Mono 
Archi 
Part 2 
U Ba 
GIS, 
Hobb: 
OU Ba 
Maso 
Surve 
O Be 
GIS \ 
Integ 
Vol. 
Q Eh 
Remo 
Nece: 
1627 
Q El 
Integ 
Acce.