John Bosco Kyalo Kiema
EFFECT OF WAVELET COMPRESSION ON THE AUTOMATIC CLASSIFICATION OF URBAN
ENVIRONMENTS USING HIGH RESOLUTION MULTISPECTRAL IMAGERY AND LASER
SCANNING DATA
J. B. K. KIEMA
Institute for Photogrammetry and Remote Sensing (IPF)
University of Karlsruhe, Germany
kiema@ipf.uni-karlsruhe.
KEY WORDS: Data Compression, Wavelets, Data fusion, Classification.
ABSTRACT
This paper examines the influence of data fusion and wavelet compression on the automatic classification of urban
environments. The principal data used is airborne Daedalus scanner imagery. Laser scanning data is introduced as an
additional channel alongside the spectral channels thus effectively fusing the local height and multispectral information.
The feature base is expanded to include both spectral (e.g., spectral signature and texture) and spatial features (e.g.,
shape, size, topology etc.). This enables the incorporation of context information into the feature extraction. A
maximum likelihood classification is then applied. It is demonstrated that the classification of urban scenes is
significantly improved by integrating multispectral and geometric datasets. The fused imagery is then systematically
compressed (channel by channel) at compression rates ranging from 5 to 100 using a wavelet-based compression
algorithm. The compressed imagery is then classified. Analysis of the results obtained indicates that a compression rate
of up to 20 can conveniently be employed without adversely affecting the segmentation results.
1 INTRODUCTION
Besides cartographic generalisation, map updating defines a basic problem in cartography that is yet to be solved in a
satisfactory manner (Báhr and Vógtle, 1999). The automatic classification of object features from remotely sensed
imagery may be viewed as a contribution towards the solution of this problem. This is of particular interest in urban
environments, especially given the high concentration of man-made features in such areas. That most of the research
effort in the automatic segmentation of geospatial imagery is currently focused on man-made features in general, and
buildings and roads in particular, testifies to this fact. However, in view of their size, structure, spectral characteristics,
as well as simple diversity, the segmentation of man-made features defines a unique problem.
Airborne laser scanning (ALS) is well suited for the production of Digital Surface Models (DSMs). The geometric
information contained in this data can be used to support the discrimination between objects that are projected higher
than the terrain (e.g., buildings, trees etc.) from those that are basically at terrain level (e.g., streets, gardens, parks etc.).
Although ALS data provides a rich source for geometric information, the use of this data alone is nonetheless, of limited
applicability in the extraction of urban objects. In order to enhance its value and thereby exploit its full potential, it is
often necessary to integrate this with other data sources e.g., multispectral data.
Data compression is routinely employed in remote sensing within the context of transferring recorded spatial data from
satellite sensors to ground stations, as well as in the ultimate storage of this data. In addition to this, the examination of
the influence of data compression techniques on the quality and further processing of satellite imagery represents an
important and contemporary research and development topic (Shiewe, 1998). In this regard, (Fritz, 1997) observes that
compression rates above a factor of 10 will be required in order to handle future high-resolution commercial sensors.
The field of wavelets has opened up new opportunities for the compression of remotely sensed data. In contrast to the
standard JPEG lossy compression technique which works best on "continuous tone" (homogeneous) data, wavelet
compression is well suited for data which is characterised by sharp spectral discontinuities. It can be argued that by
virtue of being comprised of a mixture of different spatial objects (e.g., buildings, streets, gardens, water-fountains etc.),
urban environments often result in heterogeneous data. This type of data is therefore best compressed using wavelet-
based compression schemes.
488 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.