45
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008
demonstrated for the ADS40 (Reulke et al. 2006). Based on the
sampling theorem, the optical system has to provide the
necessary spatial frequency for Superresolution.
Figure 4. Textured DEM (image data by MFC/DLR)
Object detection and tracking are methods which are not only
employed in the fields of photogrammetry and remote sensing.
Typical applications refer to the integration of optical and
distance sensors, which are applied in the automobile industry
(e.g. vehicle sensors) (Catala-Prat et al., 2008), to tracking of
objects provided by multiple camera systems for observing
tasks and to traffic monitoring (Meffert et al., 2005). Related
approaches are interesting as they also contain synchronization
aspects between datasets (beside the classic spatial co
registration).
4. DERIVATION OF REQUIREMENTS FOR FUSION
PRODUCTS
Aspects of relevance to standardization and to the quality
assessment of fusion results irrespective of which algorithm is
applied do not appear in many published papers. The following
fields of requirements can be determined:
• Requirements for utilised sensors
• Requirements for methods and quality of co
registration
• Requirements for spatial image quality (e.g. from
point spread function analyses) and
• Requirements for radiometric and colour quality (for
true colour image data).
The above requirements have to be specified and explained
more exactly for individual approaches.
As for the requirements for image and elevation data reference
can be made to the German standard ‘Requirements for the
orthophoto’ (German Institute for Standardization DIN, 2003).
As for pan-sharpening methods the requirements correspond
exemplarily to following aspects:
• The utilized sensors have to fulfil both remote sensing
tasks and true colour demands.
• The quality of co-registration refers to low-resolution
multispectral data in comparison to the high-
resolution band.
• The spatial image quality (e.g. from point spread
function analysis) has to correspond with the image
quality of the high-resolution band after the pan-
sharpening process.
• The radiometre and colour quality (for true colour
image data) has to correspondent with the
multispectral products.
The quality aspects of pan-sharpened images are described in
greater detail below.
Requirements for data fusion
• Enhancement of spatial resolution
A quality criterion for pan-sharpening methods is the
preservation of the spatial resolution of the
panchromatic image in the end product. This can be
evaluated by analysis of the point spread function by
means of distinctive image structures.
• Preservation of spectral features
Spectral features of the original low-resolution image
need to be preserved in the generated high-resolution
multispectral image in order to be in the position to
adopt e.g. classification algorithms successfully. A
modification of the colour distribution in the end
product compared to the reference image can be
roughly endorsed by comparison of the histograms of
the red, green and blue proportions of the individual
images. More differentiated assessment of the
preservation of true colour features in the original and
the pan-sharpened images can be carried out by
applying colour-distance in Lab-space (Wyszecki and
Stiles, 2000).
Another quality criterion for the pan-sharpening process
performance is the number of visible artefacts. Especially
problematic are object edges, if the red, green and blue bands
are not accurately co-registered.
5. CONCLUSIONS AND OUTLOOK
A small number of fusion methods have been well implemented
(e.g. orthophoto generation, pan-sharpening).
Data fusion is not only an issue in remote sensing research. It
has become of operational importance. Many civilian and
defence applications are enhanced through sensor and data
fusion (e.g. transportation management, tracking, and
automotive).
The quality assessment of fusion is still difficult. Corresponding
fundamental scientific research is necessary.
From our point of view, new applications based on higher levels
(feature and decision level) are to develop in the future. An
outstanding challenge is the handling of ambiguities if different
sensors come to different fusion results on decision-level.
ACKNOWLEDGEMENTS
We would like to thank Christoph Doerstel from Intergraph for
providing the DMC images and the pan-sharpened DMC
images as well as Anko Boemer from German Aerospace
Center/ Optical Information Systems for providing the images
of the 3D model and the textured 3D model. Many thanks are