Figure 37: Original image (SPOT P)

Figure 38: Original image (SPOT XS3)

Images ( Figure 37) and ( Figure 38) show a portion of the CANAL du MIDI between A and B. The north region is an urban zone, and the canal is occluded by vegetal cover, mainly noticeable on channel XS3. The database available has a maximum error of 45 meters (6 pixels). To increase the difficulty original vectors have been modified to have a maximum error of 60 meters (8 pixels); we use a search area of 10 pixels large.

Figure 39 shows the result of dynamic programming on valleys extracted from SPOT P. Two errors may be noticed in A and C, and a very important one in B. Figure 40 shows the dynamic programming on valleys extracted from SPOT XS3, there are two important errors A and B in the occluded upper part. As we used two different images to perform the extractions, we use the intermediate fusion to combine the results. The dynamic programming on the intermediate fusion ( Figure 41) is nicer; one can see only one small error in C, and noise effect in the occluded upper part.

Figure 39: ISEF valleys (Channel P)

Figure 40: ISEF valleys (Channel XS3)

Figure 41: Intermediate fusion

The snake ( Figure 43) fits the real position of the CANAL du MIDI, regarding to original vectors ( Figure 42), error C and the noise effect have been smoothed

Figure 42: Vectors from Data Base

Figure 43: Snake result