perClass Documentation
version 5.1 (31-May-2017)
 SDBANDS  Band feature extraction for spectral data

 Define bands manually:
     P=SDBANDS(DATA,'bands',{30:40 50:80})
 Sequential band definition by band size and step:
     P=SDBANDS(DATA,'size',10,'step',10)
 Cluster spectral domain with k-means:
     P=SDBANDS(DATA,'cluster',10)
 Display info on bands
     SDBANDS(P)

 INPUTS
    DATA     Data set with spectral walevengths (features express continuity)

 OUTPUTS
    P        Band feature extractor

 OPTIONS
    'bands',DEF   Define bands manually. DEF is a continuous vector of wavelength
                  indices or a cell array of such vectors
    'size',S      Define band size in number of wavelengths for sequential definition
    'step',ST     Define step between sequentially defined bands
    'cluster',N   Cluster spectral domain into N clusters with k-means
    'mean'        Default feature extraction method (band is a mean of wavelengths)
    'LDA'         Feature extraction: Fisher projection per band (see SDLDA)
    'no display'  Do not show any output

 DESCRIPTION
 SDBANDS creates band feature extractor for spectral measurements. By
 wavelength we denote individual feature of DATA, assuming that there is
 continuity (neighboring relationship) between the features. A band is a
 continuous set of wavelengths.

 Band extractor is defined in two steps, namely band definition and band
 feature extraction.  Bands may be defined manually with 'bands' option,
 sequentially with 'size' and 'step' or by clustering the spectral domain
 with 'cluster'.  Clustering algorithm defines similar groups of
 wavelengths in the spectral domain. Note, that spectral clusters may
 contain spectrally-disjoined wavelengths and, therefore, yield more
 spectral bands than the number of clusters.
 Band feature extractor is either mean of wavelengths or LDA projection
 maximizing class separability for each band.  SDBANDS also displays
 information on trained bands.

 SEE ALSO
 SDLDA, SDKMEANS