Hyperspectral Data Processing: Algorithm Design and Analysis

25.3 Spectral Feature Probabilistic Coding

The SDFC developed in Section 25.2 still follows a similar design philosophy used for SPAM and SFBC developed in Chapter 24. The spectral feature probabilistic coding presented in this section takes a completely different route to design codes for a signature vector. It replaces 1-bit threshold used by binary coding with a set of discrete values obtained from quantizing real spectral values. So, the number of bits to be used is determined by how fine the discrete values are desired to be used for encoding. Furthermore, SFPC uses the entire number of spectral bands as a block of memory to keep track of changes in the complete spectral profile of a signature vector as opposed to binary coding that uses only two or three adjacent bands as blocks of memory to capture changes in a very limited spectral range. As expected, SFPC can more accurately describe spectral characteristics of a signature vector than binary coding.

25.3.1 Arithmetic Coding

Since arithmetic coding is a well-established coding method, we refer to Rissanen (1976) and Langdon and Rissanen (1981) for details. This section briefly reviews the underlying concept of AC. In doing so, the best way is to use a simple example to illustrate how AC works. Suppose that X is a binary information source where {0,1} is the source alphabet space and probabilities of 0 and 1 given by 0.4 and 0.6, respectively. Assume that a source message S is a binary string specified by S = 01101. The ...