PCA example 2

Contents || Index

04/22/2005

Principal Component Analysis of Data Cube - Example 2.

( PCA on a reduced cube.)

Load cube

File / open Select more-lx-images / date-cubes / Enk / 128-96-100.raw
(Download these separately from http://www.nist.gov/lispix/Lx04Doc/dnld/dnload.html and put them in the images folder, or wherever you wish.) If a .rpl file is there, the parameters in the import window do not need to be initialized.
See cube image window - may appear black. Slide the slider - see different slices.

Make plot of spectra with entire cube: Data Cube / plot / sum spect for whole cube

Select peaks as separate layers

data cube / slider / layers
Select 6 or 7 layers, adjust sliders to enclose peaks.

Make new cube with summed layers. 'cube / from layers. Note that this cube has only 7 slices, rather than 101. However none of the 'counts' are lost - but summed. Note that the cube as read in uses 2 bytes, but that 'info / size and limits' shows that none of the pixels is over 1 byte. A new cube taking half of the memory could be made using ' cube / clip to byte', without loosing any data.

Do principal component analysis of the summed cube: PCA of original cube (less blank images).

data cube / PCA / 1 eigenvalues See message and PCA cube.
data cube / PCA / 2 score images See cube with score images.
Note - a blank cube image (a mean centered copy of the original cube) is made. It can be used in place of the original, or closed. To see slices, make a slider: 'slider / view slice'.

Show all the score images separately:

'image / slices to windows', take default (0 6) of all images. See 7 images being made.
shift / Size!.
Click on score image 6, 5, ... 0. Temporarily reduce size of large background window. Tile.

Scatter diagrams, color overlays can be made with these score images.