In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009 
155 
(a) Color composite 
• •V 
■ ii-'\ 
ÀŸ- 
(c) Detected changes 1 - 2 
№ 
y. Jr 
1 7 
(b) Detected changes 1-3 
f 
< 
■ .. 7 . X 
7# 
,/yr* 
(d) Detected changes 2-3 
(a) Optical image ©GoogleEarth (b) SAR image 1 
Figure 5: Construction site near "Donnersberger Briicke” 
(1: 30.03.2008,2: 22.09.2008,3: 17.03.2009) 
construction sites along the railway tracks where new residential 
and office buildings are planned. As radiometrically enhanced 
products they share a pixel spacing of 1.25 m on ground. The 
color composite (Fig. 6(a), 1:R, 2:G, 3:B) shows many colored 
regions, that help to identify the construction sites. But it is still 
impossible to interpret these changes. Fig. 6(b) indicates the 
detected changes by the curvelet approach. Many green struc 
tures stand for an increase in reflectivity over the period of one 
year. A higher reflectivity refers to new objects, e.g. walls or 
houses while the darkened regions (in red) usually refer to strong 
scatterers that have disappeared, e.g. scaffoldings. At the bot 
tom left there are sequences of green and red lines which can be 
interpreted as new buildings. One the one hand a new risen build 
ing causes a higher reflectivity (green), on the other hand it also 
causes new radar shadows (red). Some long green or red lines 
can be perceived in the middle of the image that refer to trains 
in the railway depots. Having a look at Fig. 6(c) and 6(d) much 
more small structures especially at the top right appear. Most of 
these are marked in red in Fig. 6(c) and in green in Fig. 6(d), so 
that they compensate each other over the whole year (Fig. 6(b)). 
Figure 6: Change detection (cf. Fig. 5) 
These changes are mainly found in the ’’Hirschgarten” park (see 
Fig. 5(a) at the top right) comparing the images acquired in spring 
with those acquired in fall. As these changes are restricted to nat 
ural surroundings, they supposedly refer to seasonal changes in 
the reflectivity by the tree’s growth. The blank branches in March 
cause a much higher reflectivity in the co-polarized channel than 
the leaves in September. Again the curvelet approach produces a 
change image with no single pixel disturbances. Changes in the 
underlying structures are emphasized. Unfortunately it is a diffi 
cult task to distinguish man-made changes from seasonal changes 
in the natural surrounding without a high resolution land cover 
mask. 
6 CONCLUSION 
A new approach for SAR image enhancement and change de 
tection based on the curvelet transform has been proposed and 
applied to TerraSAR-X data of the city center of Munich. As in 
put data any amplitude image can be used, for change detection 
two equally sized and co-registered images are necessary. Radar