Wavelet Compression And The Automatic Classification Of Landsat Imagery.

Abstract

Investigation of the influence of data compression techniques on the quality and further processing of satellite sensor imagery is a research and development topic which remains both current and important. In principle, this extends the scope of data compression in remote sensing beyond the classical applications of data transmission and storage. Against this background, the main objective of this study is to examine the effect of data compression on the automatic classification of Landsat imagery. The main focus of this is on the built environment. In order to obtain better segmentation results, the feature base is expanded to include both spectral features, such as spectral signature and texture, and non-spectral features such as shape, structure and topology. This expansion enables the integration of context information into the feature extraction process. The image is systematically compressed (channel by channel) at compression rates ranging from 5 to 100 using wavelet-based software. An integrated, pixel-based classification approach is then used to segment the compressed imagery. The analysis of the results obtained indicates that a compression rate of up to 20 can conveniently be used without adversely affecting the segmentation results. The suitability of wavelet compression schemes for the compression of heterogeneous data is also underlined.