Hyperspectral Data Processing: Algorithm Design and Analysis

25.1 Introduction

Spectral binary coding methods such as SPAM binary coding (1988) and spectral feature binary coding (SFBC) (Qian et al., 1996) presented in Chapter 24 are the simplest ways to characterize a spectral signature vector via a custom-designed binary code book to capture spectral variation across its spectral wavelength coverage. For example, SPAM encodes an L-dimensional signature vector as a (2L − 2)-dimensional binary code word that is composed of the first L binary values encoded as the sign of the difference between a signature component and its spectral signature mean, and additional L − 2 binary values encoded as the sign of the difference between a signature component and its corresponding signature components in its adjacent spectral bands within a signature vector. SFBC extends SPAM by including another set of additional L binary values to encode a signature vector as a (3L − 2)-dimensional binary code word where the new added set of L binary values is used to indicate whether the deviation of a spectral value in a band from the spectral signature mean of a signature vector to be encoded is greater than a threshold that is an average of the absolute differences between the value in each signature components and the spectral signature mean.

As an alternative, SPAM and SFBC can also be considered as encoders that use 1 bit to encode the current change in spectral variation and 1 bit (SPAM) or 2 bits (SFBC) to memorize the spectral changes among adjacent bands. ...