PCA example 1

Contents || Index

04/22/2005

Principal Component Analysis example 1

Load cube

File / open
Select more-lx-images / date-cubes / Enk / 128-96-100.raw, and 128-96-100.rpl
(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 matching .rpl file is not there, then open the File / Import - Export Tool, set the parameters, then use i/o / import from file.
Note that the format of the .rpl file is now text, tab delimited, as is written from a spread sheet program.

Sample rpl file:

key value
width 128
height 96
depth 101
offset 0
data-length 2
data-type signed
byte-order little-endian
record-by image ;256x96 128x96 6

Notes:

Top row has collumn labels. Need at least two (arbitrary) labels, as the .rpl file must have at least two collumns.
Values in rows are separated by one tab, and any number of spaces.
Other collumns are ignored.
Anything after a semicolon is ignored.

Lispix opens cube showing slice 0. Image of this size better seen using 'zoom / 2' , which will also get rid of the white space.

PCA of original cube (less blank images).

TOOLS / data cube / PCA / 1 eigenvalues See message and PCA cube.
TOOLS / data cube / PCA / 2 score images See cube with score images.

Use regions slider to eliminate the blank images at the beginning. I eliminated more slices than is absolutely necessary, I suppose. I cut off some at the end, just for good measure.

Notes:

The slider can be resized, as well as the plot window.
Examine regions of the plot in detail using 'view / navigator' which works for both images and plots.
PCA / of cube again.
See score cube - here zoomed x2, and slider resized. Also see Eigenvalue chart, showing most of info to be in the first few slices.

Color overlay and score plots

Extract first 3 slices into individual images:

'image / slices to windows'
Select slices: 0 2.
See three new images. Shift / resize, Tile:

The order of the images depends on when they were clicked.

They can also be sorted using 'window / sort up' and 'window / sort down'.

Scatter plots:

'TOOLS / 2 & 3 variable'.
Select the three images by clicking first on the image, and then on the appropriate button, in turn.
'RGB overlay' / make RGB overlay.
Then 'zoom / 2''.
'scatter plots ' / 2-d

Larger images have more counts in the scatter diagram, and have more colors.

Traceback:

Draw rectangle on the xy plot using 'Select / draw rectangle'.
Make mask image with 'Select /mask from rectangle' (B, below).
(The mask image serves as a record of what area of the scatter diagram was selected. The yellow rectangle is forgotten when another one is drawn.)
Click on mask image, then '2 & 3 variables / mask! '
Make traceback mask (M, below) with 'scatter plots / traceback'.

To highlight the pixels in the traceback mask:

Make an image of the cube data: 'Image / sum image from whole cube'

To overlay the traceback on the sum image:

'TOOLS / 2 & 3 variables'
Click on the sum image, then '2 & 3 variables / green/y!
Click on the traceback mask image, then on
- '2 & 3 variables / red/x!' , and then
- '2 & 3 variables / green/y!
Make the RGB overlay with 'RGB overlay / make RGB overlay'. The image data will be in green, with the highlighted areas in magenta - white.