The Luxuriant Image — an Introduction to Remote Sensing Data Processing

Abstract

The text is an introduction to remote sensing data processing by computer, and is addressed to naturalists (geographers, geologists, pedologists, agronomers, oceanographers) with no particular computer science background. Emphasis lies on computer use to process and visualize remote sensing data, and on its inherent limits and possibilities.

The nature of the remote sensing images (actually a data matrix) is first described; these images may be acquired as photographical products (transparencies) or digital products (computer tapes). The chain of events which generate digital data, from the spaceborne sensor to the user magnetic tape, and data organization on these tapes, are briefly reviewed. Data visualization is then examined in detail. The relationships between digital data and the pictures which may be obtained from them on a screen or a hardcopy are complex and somewhat surprising. It is actually impossible to see the digital “image” in its integrality. But it is possible to derive an impressive number of quite different presentations of this image. For this, the role of transfer-functions, which are of fundamental importance, are introduced and analysed, and their relationships with channel histograms are illustrated through some examples. First, the visualization of a single channel image in black-and-white is studied. Then the manners in which the screen may be coloured, and how up to three channels may be seen together (but not more) are analysed.

The notion of channel space is then introduced and some treatments which can be directly derived from the study of this space, such as derived channels, are deduced. The derived systems: Intensity-Hue-Saturation, the Tasseled Cap, Thematic Channels, and Principal Components are briefly reviewed and the notion of robust algorithms is proposed.

Radiometric corrections and geometric corrections, which are necessary to obtain workable data, are briefly explained. The text ends with a very short introduction to context-free (i.e. pixel by pixel) classification procedures.