meta data for this page

SID – Spectral Information Divergence

SID with an input and a reference image, and its argument dialog Spectral Information Divergence (SID) measures how different the spectrum of each pixel is from one or more reference spectra, by measuring the distance between the probability measures of the two spectra. It is used like SAM, with a different measure.

Inputs

SID has two inputs, an input image and a reference image. Both must have the same number of spectral bands (at least 2). The reference can be a single pixel or a small, cropped region of the image, or it can be the whole image, with the reference windows entered in SID's “roi” argument. Each window is four numbers, x y width height, as produced in Show. If you give no window, the reference is the average of the whole image.

Reference regions

Enter several windows in “roi”, one after the other, and choose “regions”:

  • average: all the regions are averaged into one reference, and the output has one band.
  • separate: each region is a separate reference, and the output has one band per window.

Output values

The closer an output value is to 0, the more similar the pixel's spectrum is to the reference spectrum. The output is floating point.

SID does not allow negative values. If the input has any, Hoku warns you and converts them to positive values (their absolute value), which might give unusual results. It may be better to convert the image first with Pack, using “linear” (called “extremes” before Hoku 3.2) and a 16-bit integer output. Check Suppress negative value warning to turn the warning off.