PCA – Principal Components Analysis

Principal Components Analysis (PCA) looks for variance in multispectral data and re-orients the axes of the data so that it is ordered by decreasing variance. At times, this can reveal difficult-to-see text, by separating the low-contrast information from the high-contrast information.

PCA has two inputs, both from the same multiband image (with at least two bands):

• roi_link: a region of interest. PCA computes the principal components from this data, so read it with a “window” around the area you care about.
• input_link: the image that the resulting PCA transform is applied to, usually the whole image.

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

There are three ways to compute the components: variance (the default), correlation and crossproducts. They can give slightly different results, which might reveal different characters.

The more spectral bands, and the larger the region of interest, the longer the computation takes.

By default, there are as many output bands as input bands, ordered by decreasing statistical variance. As a result, the low-numbered bands have the higher-contrast information. Erased text is often low-contrast information, and may be in the higher-numbered bands. The highest-numbered bands are almost always random noise. If there are N output bands, they are numbered 0 to N-1.

Enter a number M in bandsOut to output only bands 0 to M-1.

The output is floating point. To save it, use ENVI with BSQ interleave; see Multispectral Image in ENVI. Calibrating the images first may give a better result; see Read Calibrated Images.