nhanced 
sequent 
and TIN 
; dine 
ad for 
lation, 
terrain 
n fact, 
rograms 
tion to 
terrain 
'e lines 
as the 
regular 
keleton 
mpling. 
t areas 
ved by 
ng maps 
try. At 
contour 
eported 
as 50%. 
e that 
nd DTM 
manual 
quality 
be an 
  
interesting alternative for several DTM 
applications. This: can be achieved by 
automatically deriving skeleton lines from a given 
set of contours and subsequently using both data 
sets for constructing a triangulated irregular 
network (TIN) as the basis for interpolation. 
Automated Derivation of Skeleton Lines 
  
The underlying idea is to imitate manual 
interpolation on a contour map. In regular surface 
areas we interpolate directly between adjacent 
line segments. Where a group of contour lines 
spontaneously changes direction, i.e., at distinct 
ridge or drainage lines, we extract intuitively 
and approximately the relief skeleton before we 
interpolate. At peaks and pits we examine the 
neighbours of the inner contour to estimate the 
local extrema before we interpolate. 
Obviously a contour pattern carries more 
information than just elevation at a set of 
points. Although the skeleton lines of 
contours--whether determined manually or 
automatically--are only an approximation of the 
relief skeleton, better DTMs will be produced than 
without this additional information. Triangulating 
the digitized contours and skeleton lines provides 
a natural structure for this kind of data, if it 
is assured that the skeleton lines do constitute 
triangle sides (see figure 2). Some of the very 
first DTM systems in the sixties were TIN-based 
(see, e.g., [15]). Now, being able to deal with 
topology is certainly one of the aspects which 
make TIN DTMs attractive again. 
  
Fig. 2: Contours, skeleton and triangles 
Cartographic generalization as applied to most 
existing topographic maps followed a ‘comparable 
strategy: maintain the skeleton, modify the “ills. 
It can be expected, therefore, that contours from 
topographic maps are of higher fidelity at 
structural locations. This is good to remember 
when trying to extract additional information from 
the contours in order to enhance DTM quality 
without recourse to further measurements. 
et al [2] reported on two approaches under 
investigation for automatically deriving skeleton 
lines from contours: a vector and a raster 
approach. Both methods--one using aspect 
information, the other based on medial axis 
transformation--lead to a significant improvement 
of DTM fidelity. Li and Chen’s [8] research 
initiative utilizes mathematical morphology for 
this purpose, relying on global shape operators 
rather than local ones. 
Aumann 
the obvious developing 
version runs on 
For us, ILWIS 
platform. The present 
IBM-compatible PCs and offers a wealth of image 
processing and GIS tools. Its DTM component is 
raster based, converting digitized contours by 
Borgefors distance transformation and linear 
interpolation to grid data (see [6]). ILWIS 
includes digitizing software that supports manual 
digitizing of contour lines. For further 
processing, they are converted to a raster image. 
[14] is 
9577 
An attractive alternative to creating such a 
contour raster image would be to scan the existing 
contour map, followed by automated pre-processing. 
This has been a subject of research activities for 
several years, and recent publications include 
[7], [16]. 
PROCESSING STEPS 
Given contours in raster form, the series. of 
processing steps shown in figure 3 produces a DTM 
of better fidelity than one produced by 
interpolating from the contours. 
rasterized contours rasterized contours 
by point by line 
distance Dirichlet 
transformation tessellation 
ee 
extraction of 
skeleton line ‘segments 
d 
  
    
  
  
  
  
  
  
  
  
  
  
  
  
> rare ee EE 
Y !gaps closing (a)! 
, L--- —-+ 
    
  
— —— — 
| gaps closing (b)! 
J 
Hs ee oe en 
| 
ET 
line compression 
and smoothing 
height assignment 
rasterizing 
rasterized skeleton 
  
  
  
  
  
   
    
      
  
  
r 
[triangulation] 
| | | 
[arid pri contours /slope & aspect/ 
  
| 
   
  
   
   
volumes 
and other 
applications 
   
relief shading 
  
Fig. 3: Automated skeleton extraction and 
triangulation 
In the folloving, the main features of the process 
are outlined; details about the algorithms and the 
developed experimental software are given in [10].