19.1 Introduction

Because of significantly improved spectral and spatial resolution resulting from recent advanced remote sensing instruments many subtle substances such as rare minerals, special species, small objects etc. can now be uncovered and diagnosed by custom-designed data processing techniques such as feature extraction for exploitation. However, this benefit also comes at a price, that is, how to process enormous data volumes without compromising desired information for data processing, specifically, how to compress data while preserving vital information for future information retrieval and data processing. Apparently, this heavily depends on the data to be processed. Different data are acquired for various applications; thus, they require specific processing techniques. This chapter investigates hyperspectral data compression from an information point of view, referred to as hyperspectral information compression.

Before proceeding we need to make a distinction between information compression and data compression. Let us consider the following example. Assume that a document such as a newspaper is represented by a binary image with 0 corresponding to letters and 1 being assigned to background so that the document can be read by black letters in white background as shown in Figure 19.1(a).

Figure 19.1 An example of a print from a newspaper.

img

Now, if we perform a lossless ...

Get Hyperspectral Data Processing: Algorithm Design and Analysis now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.