Jamet - 9 
4.2. Weaknesses of automatic methods. 
Though we haven't experimented it yet, the main breakthrough for the automation of the 
contour line production appears to be the efficiency of the post editing process. This 
step includes: 
- the detection of the errors in the result of the automatic phase; 
- their actual correction, either through DEM correction, or through contour editing. 
The efficiency of the correction itself relies on the tools given to the operator. The 
weigh of this task should be proportional to the corrected area. However, if necessary, 
contour line editing may result heavier than DEM correction (in dense contour areas), 
since it will induce a control on the consistency (continuity of the contours, absence of 
crossings...). 
The detection phase raises more questions. Detecting errors in cartographic data (by 
comparing them with an aerial stereo pair) is well known to be a very tedious work that 
can be nearly as time consuming as a primary acquisition. In the present case, the 
operator will profit by a natural help from the terrain shapes yielded by the automatic 
method. Errors generally show shape aberrations on the terrain model (spurious local 
extrema, or distortions due to vegetation) that yield strange contour line shapes. We 
hope that this will be enough for the operator to identify the errors at a rather small 
display scale. If not, we'll probably have to develop some diagnostic tool to point out 
automatically the errors to the operator. 
4.3. Discussion 
Introducing automatic tasks in cartographic data capture processes can certainly bring 
some productivity improvement. However, in the short term, these improvements will 
keep low. The productivity of semi automatic techniques is bounded by the time 
required to manage the interactions between the operator and the computer and by the 
time spent in controlling the logical structure of the representation. This time increases 
with the complexity of the objects to be captured, and with the complexity of the 
available capture tools. Under the simple assumption that the machine should only 
relieve the operator from that tedious part of his work which is the capture of the 
geometrical characteristics of the objets, no time will be saved on too small objects. 
If there seem to be more hope in automatic techniques, it's under the assumption that the 
logical structure of the data is also managed by theses processes. The potential success