meta data for this page

MNF – Minimum Noise Fraction

MNF with three inputs, and its argument dialog Minimum Noise Fraction (MNF) looks for differences in variance in multispectral data, like a PCA transform. The difference is that MNF first whitens the data, using the statistics of a noise region, before computing the Principal Components Analysis transform.

Three inputs

MNF has three inputs, all from the same multiband image (with at least two bands), but of different regions:

  • noise_link: a reasonably uniform region of the image, where the noise statistics are measured. They are used to whiten the data.
  • roi_link: a region of interest, where the transform is computed from the whitened data.
  • input_link: the image the transform is applied to, usually the whole image.

The three images must have the same number of bands, but they can be different sizes.

The more spectral bands, and the larger the region of interest, the longer the computation takes. The results can vary greatly as you change the roi and noise regions.

Arguments

Argument What it does Default
analysis variance subtracts the mean of the data when computing its covariance; crossproducts does not, as in PCA variance
preprocess whiten also whitens the input image before the transform is applied none
bandsOut The number of output bands, M. Bands 0 to M-1 are output as many as the input

Output

The output is floating point, with as many bands as the input unless you set bandsOut. Because the data is whitened before the transform is calculated, the output bands tend to decrease in image quality: the lower-numbered bands are more likely to contain useful text.