Hi, there,
I come back again because my coworker reminded me to followup my question. (even I though that is weird to ask and answer the same question by myself. and I hate the answer that I found.)
ok, then....
Following the statistical definition, if I would like to get the brick's covariance I have to add -dmean option before the covariance calculation.
and then I can got the similar result to that in Matlab (just scaling problem between them, such as 2.985945x10^8).
and last time, I said I got different result is because I used 4-byte precission in my input data. and then the result turns out a weird way. so maybe the precission issue of calculation in eigendecomposion is very critical, or maybe my data is under ill condition.
that is the story....
and I hate this finding because I have to redo the processiong in my data in a time consuming way.
and I am also curious about the precission issue. does more precissions get better result? why don't we use 20 bytes or more to describe our data?
just complain...