In: Wagner W„ Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
631
the landscape shows a disconnected area. Figure 8 (a)~(d) show
the reference DEM and the filtered DEM by three filters
respectively. We can see that the erosions and smoothness were
produced at the disconnected areas which generated by
TerraScan and ASF. The disconnected area generated by
DDASF is closer to reference DEM.
Erosions due to over-filtering
Figure 8. Filtering results of Sample 23 by three different
filtering methods.
3.2 Filtering results of Sample 53
Sample 53 is located at suburb areas and contains many features
of discontinuity preservation. There are cliffs with large
elevation-difference over the lower-right landscape and terraced
filed over the upper-left landscape. Figure 9 (a)~(d) show the
reference DEM and the filtered DEM by three filters
respectively. We can see that there is less difference between
the DEMs produced by TerraScan and DDASF while DEM
generated by ASF reveals a quite difference at the discontinued
areas. This is due to the over-filtering situation. Figure 10
represents the differences between reference DEM and DEMs
which generated by the different filtering methods. From Figure
10(b) and (c), DDASF shows an improved result when a
directional window is implemented. Generally DDASF can
generate the closest DEM over the discontinued areas in this
case, since the black(represents larger difference) areas in
Figure 10(c) are thinner than others.
(a) Reference DEM (b) DEM by TerraScan
(c) DEM by ASF (d) DEM by DDASF
Figure 9. Filtering results of Sample 53 by three different
filtering methods.
(a) Difference between (b) Difference between ASF
TerraScan DEM and DEM and Reference DEM
Reference DEM
Figure 10. Difference between reference DEM and generated
DEM
Besides, the classification accuracy and RMSE are also
calculated to evaluate the three methods. It is noticed that a
good classification accuracy on a sample does not stand for a
well RESM on the corresponding sample, too. For example the
total error of Sample 53 produced by TerraScan is minimum