i.pca could be wrong

Agustin Lobo lobo at Jasper.Stanford.EDU
Fri Mar 12 13:27:11 EST 1993


I have reported the following on bug.report.sh for grass4.0:

Bug Report: the correlation matrix obtained with r.covar -r on the output
of i.pca is:
 
1.000000 0.556467 0.430150 
0.556467 1.000000 0.851286 
0.430150 0.851286 1.000000 
 
which appears to me to imply that the pca was wrong: the correlation matrix
of pc's has to be, by definition, with only 1 in the diagonal an 0 anywhere else. A run involving m.eigensystem yields instead, results which once analyzed withr.covar -r are consistent with the teory. This could be a very important bug, for pca is a previous step in many image processing methods.
Jasper.Stanford.EDU% mailtool &

Agus



More information about the grass-user mailing list