algorithm is tested using ISNAP. Examples are also given, 
which demonstrate the potential of the proposed algorithm. 
Our discussion will restrict on TIN DTMs. Contour lines 
are used as constraints in the triangulation process. 
2. A CONCEPT FOR THE GENERALIZATION OF 
RELIEF REPRESENTATION 
While contour lines are the most comprehensive form of 
terrain relief representation in a 2D analogue environment, 
the digital terrain model is the approach for representing 
terrain relief in a digital environment due to its advantage in 
computer analysis. Terrain relief generalization hence can 
be regarded as an issue of DTM generalization, and 
(conceptually) contours can be seen as one of the graphic 
representational forms of a DTM in a GIS context. Thus 
generalization of DTMs and generalization of contour lines 
fall within the frameworks of database generalization and 
view generalization respectively (Peng et al., 1996). This 
relationship can be further demonstrated by the fact that 
contour lines of any interval can (and should) be derived 
from a (good) DTM, and the fact that generalization of 
contour lines is restricted to the graphic aspect of 
generalization (Bos, 1984), except for the selection of 
contour line interval which is associated to the spatial 
properties of terrain surfaces, apart from other aspects such 
as map scale and usages. 
DTM generalization aims at reducing the spatial (relief) 
resolution of a source DTM to arrive at a more abstracted 
relief model. The factors that affect the selection of a proper 
resolution for an application may include, for instance, the 
purpose, the relevance of small details, accuracy 
requirement, processing time, data storage space, hardware 
and software limits. It is important to stress that although it 
is true that in general a more abstracted relief model is also 
more smooth and less accurate, smoothing or compression 
operation alone does not, in general, provide good 
generalization result. The key aspect is that while local and 
irrelevant relief details disappear, the skeleton information 
representing the characteristics of the terrain surface should 
be maintained as much as necessary. From this point of 
view, both DTM filtering (Loon, 1978, Zoraster et al., 
1984) and DTM compression (Gottschalk, 1972, Heller, 
1990) are not adequate approaches. However, they can be 
improved by introducing skeleton information as a 
constraint in the generalization process. 
3. AN APPROACH TO THE PROBLEM OF 
GENERALIZATION 
Known approaches to the problem of relief generalization 
can be categorized into three groups, namely: (1) DTM 
filtering (Loon, 1978, Zoraster et al, 1984), (2) DTM 
compression (Gottschalk, 1972, Heller, 1990), and (3) 
structure or skeleton line generalization (Wu, 1981, Yoeli, 
1990, Wolf, 1988, Weibel, 1989). Weibel (1992) evaluated 
these three types of methods and pointed out that global 
filtering (or DTM filtering) achieves a smoothing effect by 
eliminating high frequencies from the source DTM while 
keeping the number of points in the model unchanged. 
Selective filtering (or DTM compression) selects a subset of 
points from the source DTM to approximate the original 
surface with a user-specified accuracy. While both 
approaches are employed for minor scale reductions, DTM 
filtering is intended to be used in topography with smooth 
forms, and DTM compression is meant to be applied to 
terrain of any complexity. Heuristic generalization (or 
structure line generalization) directly generalizes the 
structure lines of the terrain surface through individual 
generalization operators (i.e., selection, simplification, 
combination, displacement, and emphasis), and 
reconstructing the target DTM through interpolation from 
the generalized structure. It is intended for use in rugged 
terrain and is the only approach that includes the 
    
  
    
fundamental transformations (i.e., combination and 
displacement) required to accomplish major scale 
reductions (Weibel, 1992). 
Other sources 
c) 
f Constraint DTM 
Compression 
  
  
  
  
  
    
Intermediate DTM 
Constraint 
  
  
  
DTM filtering 
    
Structural lines 
Generalization 
€— 
| 
Jen 
Generalized skeleton, 
  
  
  
  
   
   
Intermediate DTM | 
  
| 
e) 
  
  
Additional information 
» Verification 
Generalized DTM 
Figure 1 The proposed generalization process. 
  
  
  
  
  
In fact, these three generalization approaches emphasize on 
the different aspects of generalization: DTM filtering 
smooths the surface but does not reduce the data volume, 
DTM compression reduces the data volume but does not 
necessarily lead to a more abstracted surface, and structure 
line generalization deals with skeleton transformation but 
ignores other properties not shown in the skeleton. Hence, 
an approach combining these three methods may lead to a 
more comprehensive solution: a) extracting the skeleton 
from the source DTM or from other sources; b) generalizing 
the skeleton through structure line generalization; c) 
creating the first intermediate DTM by applying DTM 
compression to the source DTM and using the generalized 
skeleton as a constraint (e.g., instead of using the non- 
collinear points on the convex hull and “significant 
extremes", the generalized skeleton can be used as the 
initial set of points); d) creating the second intermediate 
650 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996