Hyperspectral Data Processing: Algorithm Design and Analysis

25.2 Spectral Derivative Feature Coding

This section presents a new coding method, referred to as spectral derivative feature coding (SDFC), that extends and improves both SPAM and SFBC in the sense of signature characterization. The idea is derived from texture analysis based on a feature coding method developed by Hong et al. (2002). Instead of characterizing texture features according to spatial correlation among nine pixels in a 3 × 3 window, SDFC converts image-based texture features to spectral derivative features by looking into spectral variations among three adjacent bands within a signature vector. In doing so, it reinterprets SPAM and SFBC as 1-bit binary encoders using various numbers of bits to store memory as follows.

25.2.1 Re-interpretation of SPAM and SFBC

Using the sample mean of spectral value of all bands in a spectral signature vector specified by (23.1), with L being the total number of spectral bands, we can implement SPAM as a 1-bit binary encoder with 1-bit memory to encode s as follows.

For each lth band, we encode the s_l by the following code word, denoted by :

(25.1)

and

(25.2)

Concatenating in (25.1) with in (25.2) results in a code word for s_l, denoted ...