Fischler, 
known 
research laboratories 
McKeown, Yee). The 
hierarchically: 
Pieces of linear elements, extracted by contrast or 
line detectors are progressively merged into more 
complex objects under topologic and morphologic 
constraints (connectivity, alignment...). 
These methods have missed their goal, mainly 
because they only deal with geometric properties and 
they lack higher level knowledge about objects 
semantic and real world organisation. 
This high level knowledge gives to the human visual 
cortex the ability to identify, at first sight, roads or 
railways in an image. 
The snake concept take advantage of this ability in 
the following way: 
A feature, detected by an operator, is approximately 
drawn on the screen. This initial polygonal line is 
automatically moved (snake analogy) until it meets 
some optimal condition, depending on actual position 
of the real feature in the raster image. 
Feature identification is human driven and geometric 
accuracy is ensured by the computer. 
(ref. 
best 
Groch, 
proceed 
This concept has been implemented as a computer 
assisted interactive graphic editor. 
Every polygonal element can be automatically 
adjusted to the nearer feature in the underlying 
image. 
The algorithm is based on dynamic programming 
techniques. The vertices of the polygonal line are 
locally moved until radiometric properties along the 
line realize the best fit to a given model of linear 
feature. 
  
initial hand drawn line 
  
automatic adjustment 
344 
3.3 Topological relationship 
A topological graph is made of arcs (or lines), nodes 
and polygons. 
An arc describes a single linear element (without 
crossing). 
A node defines connection between two or more 
arcs. 
A polygon is a sorted list of connected arcs defining a 
single closed area. 
Each arc contains references to the polygon on its 
right and the polygon on its left. So, any spatial 
relation between polygons can be easily computed. 
a2 
P3 
P2 
EL 
ab 
P4 
topological graph 
The interest of topologically structured data is 
obvious for geographic information processing: lt 
enables efficient spatial reasoning and filtering. 
Up to date geographic information systems make use 
of topologic data files. 
From a data producer point of view, topological 
relationship makes syntactic data control easier: 
Network connectivity, land cover completion and 
areas boundaries consistency can be automatically 
checked. 
For productivity sake, relation edition should not 
burden operators with an additional task. 
We have integrated a topologic structuring capability 
in our graphic editor: The land cover graph is 
automatically generated, so are  "X-crossing" 
connections within networks. “T-crossing” 
connections still have to be manually edited. 
Toe 
X-crossing connection 
Lo 
T-crossing connection