Figure 5: Line pixels (black) and no-line pixels (white) ex- 
tracted from intensity. 
  
  
  
  
  
Figure 6: Line pixels (black) and no-line pixels (white) ex- 
tracted from intensity and coherence. 
cess of the ribbon snake-based extraction: Only by opti- 
mizing from the given ends inwards it was possible to track 
the pipeline avoiding to get stuck to other lines with high 
posterior odds. 
5 CONCLUSIONS 
The fusion of SAR intensity and interferometric SAR co- 
herence data for line extraction using a Bayesian approach 
has been introduced. The new method applies an MRF 
model to suppress speckle-related line gaps. Results are 
line pixels with line directions and posterior odds showing 
  
Figure 8: Pipelines and roads extracted with ziplock snakes 
from posterior odds superimposed to SAR magnitude. 
the strength of the line state in comparison with the no-line 
state. 
The posterior odds contain line information from both data 
sources. They can be used for further object extraction. 
Here, they have been input to a ziplock snake-based ex- 
traction of pipelines. Though, it would be desirable to 
conduct the extraction automatically, interactive process- 
ing was necessary to achieve reliable results. The rea- 
sons for difficulties of automatic processing are twofold. 
(1) Despite of the speckle suppressing effect of the MRF 
model, the posterior odds have a high noise level. They 
contain many structures which are caused by other objects 
than pipelines, e.g. by terrain features, which confuse auto- 
matic extraction. (2) The object model of the ziplock-snake 
536 Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 
Coc 
anc 
plic 
Fua 
tion 
Ger 
Gib 
age 
Inte 
Goc 
Spe 
and 
Spr 
Hell 
mit 
Deu