In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Vol. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009
Metadata
GCPs Imagery data j
Given/Externally
Computed RPCs
Image Pre-processing
Triangulation (Tie Point Measurement & Block Adjustment)
Image Matching
Quasi-Epipolar Image
Generation
DSM/DTM
Generation
Orthophoto
Generation
Object Measurement
(Manual / Semi-automatic: Mono
plotting / Stereoscopic)
I
CC Modeler (Construct
3D Models & Texture
Mapping)
3D Database, DTMs. Orthophotos, Vector
Data, etc.
Figure 2. SAT-PP workflow © Chair of Photogrammetry and
Remote Sensing ETH Zurich.
4.2 Preprocessing of the satellite data
Before processing the VHR imagery a contrast enhancement is
executed as this leads to a more reliable image matching.
Especially between images of the same area but taken at
different dates from different orbits large radiometric
dissimilarities can occur due to different illumination and
atmospheric conditions, leading to poor matching results. To
enhance the contrast for each image individually and to equalize
the radiometric differences between the imagery, a Wallis filter
was applied (Wallis, 1976).
The general form of a Wallis filter is given by:
g W (x,y) = g(x,y)*r ]+ r 0
(1)
r \
cs h
(2)
r 0 = bm h + (1 -b-r x )m g
(3)
with g w (x,y) and g(x,y) = filtered and original image
m c and s g = original mean and standard deviation values
m h and s h = target value for mean and standard deviation
c and b = contrast expansion and brightness forcing c lc
The Wallis filter performs a non linear, locally adaptive contrast
enhancement. Actually a large kernel divides the image in
different sections and within each section the local contrast is
optimized. Applying a Wallis filter on the original images does
not only result in an enhancement and sharpening of texture
patterns in areas of low contrast and equal overall contrast but
normalizes also the radiometry, especially between images
taken at different dates. The effect of radiometric enhancement
of very high resolution satellite imagery is illustrated in figure 3
& 4. The Wallis filter enhances existing texture patterns,
leading to optimization of the contrast in shadow areas. Note
that in the shadow rich areas axis-aligned artefacts are
introduced due to the Wallis filtering.
Figure 3. Extract of original 11-bit Ikonos image, illustrating an
area with high buildings. There is very little contrast within the
shadow areas, leading to mismatches during the image matching
process.
Figure 4. Extract of Wallis-filtered 11-bit Ikonos image. The
radiometric filter enhances the existing texture patterns locally,
leading to optimization of the contrast in the shadow areas.
Also an adaptive smoothing filter is applied to reduce image
noise while sharpening edges. As noise is an important data-
source for mismatches, reducing it further improves the quality
of the surface model.
Next to the radiometric enhancement a method for geometric
normalization was devised. The Ikonos 2002 stereo couple is
epipolar projected and suitable for stereo applications. As the
2005 Ikonos image is taken from a different orbit, the images
are displaced to each other and the internal geometry will be
slightly different because of the different scan direction.
Geometric normalization of the 2005 Ikonos image with the
